1{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "7dc471c38e2b7b3349fd80b1", "attempt_id": "14fab3a71a89", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:44.789151Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 64, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 67.28591700084507, "ttft_ms": 28.43649999704212, "decode_ms": 46.75175005104393}, "total_wall_ms": 149.4781249202788, "ttft_ms": 28.43649999704212, "input_usage": {"receiver_prompt_tokens": 64}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6596116480, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6221848952, "mlx_cache_bytes": 52763976, "process_rss_bytes": 6596116480}, "peak_prefill_memory": null, "peak_decode_memory": 6221848952, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}2{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "1d905b8a74595c32f476900e", "attempt_id": "1c6956028bb4", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:44.969183Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 70, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 58.32883401308209, "ttft_ms": 19.682042067870498, "decode_ms": 20.957999979145825}, "total_wall_ms": 99.59383402019739, "ttft_ms": 19.682042067870498, "input_usage": {"receiver_prompt_tokens": 70}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6596755456, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6243967624, "mlx_cache_bytes": 117202248, "process_rss_bytes": 6596755456}, "peak_prefill_memory": null, "peak_decode_memory": 6243967624, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}3{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f37452a7ec7d3067f51b5c24", "attempt_id": "79b1670bf9b1", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:45.091440Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 70, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 61.44025002140552, "ttft_ms": 24.80645792093128, "decode_ms": 26.245958055369556}, "total_wall_ms": 113.10295900329947, "ttft_ms": 24.80645792093128, "input_usage": {"receiver_prompt_tokens": 70}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6596804608, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6243967624, "mlx_cache_bytes": 117202248, "process_rss_bytes": 6596804608}, "peak_prefill_memory": null, "peak_decode_memory": 6243967624, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}4{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f9edb4468f6549f800cb524f", "attempt_id": "ee5214112780", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:45.225288Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 67, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 67.30499991681427, "ttft_ms": 32.58833289146423, "decode_ms": 36.201750044710934}, "total_wall_ms": 136.73595804721117, "ttft_ms": 32.58833289146423, "input_usage": {"receiver_prompt_tokens": 67}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6596853760, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6245671560, "mlx_cache_bytes": 162225480, "process_rss_bytes": 6596853760}, "peak_prefill_memory": null, "peak_decode_memory": 6245671560, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}5{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c0871b617f7308d16b328999", "attempt_id": "fae12c47b397", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:45.385364Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 75, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 52.22470802254975, "ttft_ms": 33.77595904748887, "decode_ms": 42.92874992825091}, "total_wall_ms": 129.70216700341552, "ttft_ms": 33.77595904748887, "input_usage": {"receiver_prompt_tokens": 75}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6596935680, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6248243848, "mlx_cache_bytes": 229399880, "process_rss_bytes": 6596935680}, "peak_prefill_memory": null, "peak_decode_memory": 6248243848, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}6{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "464bd186ad7ef93f726fca6f", "attempt_id": "0435daa0a044", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:45.539311Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 73, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 40.481833973899484, "ttft_ms": 40.42075003962964, "decode_ms": 27.114708092994988}, "total_wall_ms": 108.66475000511855, "ttft_ms": 40.42075003962964, "input_usage": {"receiver_prompt_tokens": 73}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6596968448, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6245720712, "mlx_cache_bytes": 273931592, "process_rss_bytes": 6596968448}, "peak_prefill_memory": null, "peak_decode_memory": 6245720712, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}7{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9994f97a49ef7eaa65bab903", "attempt_id": "1d797d30b8c8", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:45.671539Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 67, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 68.87345795985311, "ttft_ms": 45.527375070378184, "decode_ms": 43.627374921925366}, "total_wall_ms": 158.86070800479501, "ttft_ms": 45.527375070378184, "input_usage": {"receiver_prompt_tokens": 67}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6596968448, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6245016200, "mlx_cache_bytes": 273931592, "process_rss_bytes": 6596968448}, "peak_prefill_memory": null, "peak_decode_memory": 6245016200, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}8{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "43f1a24e11e8b198cbef5618", "attempt_id": "49aff4d1d2ff", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:45.852745Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 65, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0037919962778687477, "decode_payload_ms": 0.0, "receiver_prefill_ms": 39.54058396629989, "ttft_ms": 28.165541007183492, "decode_ms": 48.47637505736202}, "total_wall_ms": 116.93983396980911, "ttft_ms": 28.165541007183492, "input_usage": {"receiver_prompt_tokens": 65}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6597197824, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6222143868, "mlx_cache_bytes": 299294280, "process_rss_bytes": 6597197824}, "peak_prefill_memory": null, "peak_decode_memory": 6222143868, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}9{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "11825e5d5dd2e833533d1165", "attempt_id": "3c1154222a5d", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:45.992502Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2045, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 432.08558403421193, "ttft_ms": 54.78208290878683, "decode_ms": 402.40779099985957}, "total_wall_ms": 892.7009999752045, "ttft_ms": 54.78208290878683, "input_usage": {"receiver_prompt_tokens": 2045}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6597492736, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7061651504, "mlx_cache_bytes": 1280966200, "process_rss_bytes": 6597492736}, "peak_prefill_memory": null, "peak_decode_memory": 7061651504, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}10{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3633111fe7fdc73136eda5d9", "attempt_id": "0df61f4443d6", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:46.908866Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2051, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 539.7673329571262, "ttft_ms": 38.68245903868228, "decode_ms": 330.514750094153}, "total_wall_ms": 912.8277080599219, "ttft_ms": 38.68245903868228, "input_usage": {"receiver_prompt_tokens": 2051}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6598033408, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054786612, "mlx_cache_bytes": 2087045696, "process_rss_bytes": 6598033408}, "peak_prefill_memory": null, "peak_decode_memory": 7054786612, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}11{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c9ac4078f6630b453fba0e5e", "attempt_id": "3807587f966d", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:47.844390Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2051, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 542.4670829670504, "ttft_ms": 48.275374923832715, "decode_ms": 218.9682919997722}, "total_wall_ms": 812.990834005177, "ttft_ms": 48.275374923832715, "input_usage": {"receiver_prompt_tokens": 2051}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6598049792, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054786612, "mlx_cache_bytes": 2087045696, "process_rss_bytes": 6598049792}, "peak_prefill_memory": null, "peak_decode_memory": 7054786612, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}12{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "acbf2d57d99baa3bb3291022", "attempt_id": "ff4ce199181e", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:48.680247Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2048, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 473.96037506405264, "ttft_ms": 42.49712498858571, "decode_ms": 155.7709170738235}, "total_wall_ms": 675.7320000324398, "ttft_ms": 42.49712498858571, "input_usage": {"receiver_prompt_tokens": 2048}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 6598082560, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7063175232, "mlx_cache_bytes": 2087045696, "process_rss_bytes": 6598082560}, "peak_prefill_memory": null, "peak_decode_memory": 7063175232, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}13{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5ab2626e677fad777c044d90", "attempt_id": "48c57eec76c8", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:49.377114Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2056, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 907.967249979265, "ttft_ms": 39.17491703759879, "decode_ms": 261.24666596297175}, "total_wall_ms": 1211.6929580224678, "ttft_ms": 39.17491703759879, "input_usage": {"receiver_prompt_tokens": 2056}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6599262208, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054794796, "mlx_cache_bytes": 2092188669, "process_rss_bytes": 6599262208}, "peak_prefill_memory": null, "peak_decode_memory": 7054794796, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}14{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a34269d1bed9a1a8e01f5409", "attempt_id": "5cb8306ef833", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:50.610885Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2054, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 585.719499969855, "ttft_ms": 49.6917919954285, "decode_ms": 364.57666708156466}, "total_wall_ms": 1003.0161670874804, "ttft_ms": 49.6917919954285, "input_usage": {"receiver_prompt_tokens": 2054}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6599442432, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054792797, "mlx_cache_bytes": 2095285245, "process_rss_bytes": 6599442432}, "peak_prefill_memory": null, "peak_decode_memory": 7054792797, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "SP-38G-4217", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "e643c21a78559b96fd65325d9ce779b7b8fd6a47c53cf7c2cda1efa8412458c4"}15{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "fc45c178aa497c4f2856a0ea", "attempt_id": "55e138743342", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:51.634835Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2048, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 491.39533401466906, "ttft_ms": 49.010332906618714, "decode_ms": 45.80304108094424}, "total_wall_ms": 589.230666984804, "ttft_ms": 49.010332906618714, "input_usage": {"receiver_prompt_tokens": 2048}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599409664, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7063175232, "mlx_cache_bytes": 2095285245, "process_rss_bytes": 6599409664}, "peak_prefill_memory": null, "peak_decode_memory": 7063175232, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}16{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "520ca0673d257e517a5c275d", "attempt_id": "c34d0296c49e", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:52.245014Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2046, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.025416957214474678, "decode_payload_ms": 0.0, "receiver_prefill_ms": 511.2538749817759, "ttft_ms": 48.41466702055186, "decode_ms": 39.59991701412946}, "total_wall_ms": 602.9183330247179, "ttft_ms": 48.41466702055186, "input_usage": {"receiver_prompt_tokens": 2046}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599426048, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7062388800, "mlx_cache_bytes": 2656535549, "process_rss_bytes": 6599426048}, "peak_prefill_memory": null, "peak_decode_memory": 7062388800, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}17{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "906fb78d7fdd8a3d40b151e6", "attempt_id": "24501095d4d9", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:52.870086Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 211, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 85.79641696996987, "ttft_ms": 45.009916997514665, "decode_ms": 44.92799995932728}, "total_wall_ms": 176.73458298668265, "ttft_ms": 45.009916997514665, "input_usage": {"receiver_prompt_tokens": 211}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599589888, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6351512688, "mlx_cache_bytes": 2829091837, "process_rss_bytes": 6599589888}, "peak_prefill_memory": null, "peak_decode_memory": 6351512688, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}18{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6f069850606433588d01f4e0", "attempt_id": "c31134f6f532", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:53.068486Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 217, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 63.218874973244965, "ttft_ms": 24.810667033307254, "decode_ms": 34.904666943475604}, "total_wall_ms": 123.8859590375796, "ttft_ms": 24.810667033307254, "input_usage": {"receiver_prompt_tokens": 217}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599737344, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6354969712, "mlx_cache_bytes": 2997617661, "process_rss_bytes": 6599737344}, "peak_prefill_memory": null, "peak_decode_memory": 6354969712, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}19{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "404c6444eaa7eb6f037de747", "attempt_id": "58cb78b10461", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:53.212599Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 217, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 105.49899993930012, "ttft_ms": 50.18529191147536, "decode_ms": 29.159959056414664}, "total_wall_ms": 185.97908306401223, "ttft_ms": 50.18529191147536, "input_usage": {"receiver_prompt_tokens": 217}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599753728, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6354936944, "mlx_cache_bytes": 2997732349, "process_rss_bytes": 6599753728}, "peak_prefill_memory": null, "peak_decode_memory": 6354936944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}20{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f901d2b005bb9fd0a72aaa1b", "attempt_id": "be966da27525", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:53.421240Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 214, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 93.52445893455297, "ttft_ms": 44.21774996444583, "decode_ms": 34.86170805990696}, "total_wall_ms": 173.7389590125531, "ttft_ms": 44.21774996444583, "input_usage": {"receiver_prompt_tokens": 214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599802880, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6353364080, "mlx_cache_bytes": 3114910717, "process_rss_bytes": 6599802880}, "peak_prefill_memory": null, "peak_decode_memory": 6353364080, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}21{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a161f0e343ce3de65a917bbf", "attempt_id": "96b92663c87f", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:53.617102Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 222, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 90.69304203148931, "ttft_ms": 40.12424999382347, "decode_ms": 19.014084013178945}, "total_wall_ms": 150.86395805701613, "ttft_ms": 40.12424999382347, "input_usage": {"receiver_prompt_tokens": 222}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599950336, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6359639152, "mlx_cache_bytes": 3290874877, "process_rss_bytes": 6599950336}, "peak_prefill_memory": null, "peak_decode_memory": 6359639152, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}22{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ded82344b0d28af77a17b2a7", "attempt_id": "8f7189ac5427", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:53.788710Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 220, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 87.39187498576939, "ttft_ms": 40.889833006076515, "decode_ms": 30.488165910355747}, "total_wall_ms": 160.004667006433, "ttft_ms": 40.889833006076515, "input_usage": {"receiver_prompt_tokens": 220}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599983104, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6368421024, "mlx_cache_bytes": 3423044605, "process_rss_bytes": 6599983104}, "peak_prefill_memory": null, "peak_decode_memory": 6368421024, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}23{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ea3b8f591da6e3155447d060", "attempt_id": "afc053260c19", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:53.969958Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 214, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 105.75333295855671, "ttft_ms": 52.02795809600502, "decode_ms": 43.225124944001436}, "total_wall_ms": 202.0414579892531, "ttft_ms": 52.02795809600502, "input_usage": {"receiver_prompt_tokens": 214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599999488, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6353364080, "mlx_cache_bytes": 3423044605, "process_rss_bytes": 6599999488}, "peak_prefill_memory": null, "peak_decode_memory": 6353364080, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}24{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "83b17fd24c578f699db39b3a", "attempt_id": "1f3b6c9718a3", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:54.193435Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 212, "message_tokens": 132, "message_utf8_bytes": 699, "cache_tensor_bytes": null, "serialized_payload_bytes": 699, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.3587919529527426, "decode_payload_ms": 0.0, "receiver_prefill_ms": 97.247791942209, "ttft_ms": 36.42241598572582, "decode_ms": 36.45654104184359}, "total_wall_ms": 171.1676249979064, "ttft_ms": 36.42241598572582, "input_usage": {"receiver_prompt_tokens": 212}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600048640, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6351889520, "mlx_cache_bytes": 3539043325, "process_rss_bytes": 6600048640}, "peak_prefill_memory": null, "peak_decode_memory": 6351889520, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0b02f3343692bbb81db58494f929605d06dd28b19633a0be78255fe46e31eef8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}25{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9b506313b76f2583b6cb80ba", "attempt_id": "ec96216e8079", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:54.388890Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 595, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 154.8532920423895, "ttft_ms": 47.15045797638595, "decode_ms": 48.25195798184723}, "total_wall_ms": 251.66629196610302, "ttft_ms": 47.15045797638595, "input_usage": {"receiver_prompt_tokens": 595}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600310784, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6566965336, "mlx_cache_bytes": 3929310205, "process_rss_bytes": 6600310784}, "peak_prefill_memory": null, "peak_decode_memory": 6566965336, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}26{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9a57c4eb2ccc504dc3633552", "attempt_id": "597ce72eb6d4", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:54.663487Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 601, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 181.21874995995313, "ttft_ms": 44.996040989644825, "decode_ms": 329.73024994134903}, "total_wall_ms": 557.47912498191, "ttft_ms": 44.996040989644825, "input_usage": {"receiver_prompt_tokens": 601}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6600441856, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567653464, "mlx_cache_bytes": 4279665661, "process_rss_bytes": 6600441856}, "peak_prefill_memory": null, "peak_decode_memory": 6567653464, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C in Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "3491362e6ed6bb557ea2c25e29286d6fdbd0a0a36995d38446eddfd8450e097c"}27{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "4adda635e44bb416765ff4c4", "attempt_id": "11c2f6904d48", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:55.242397Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 601, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 171.44183290656656, "ttft_ms": 42.17550007160753, "decode_ms": 41.338249924592674}, "total_wall_ms": 256.42637501005083, "ttft_ms": 42.17550007160753, "input_usage": {"receiver_prompt_tokens": 601}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600491008, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567653464, "mlx_cache_bytes": 4279665661, "process_rss_bytes": 6600491008}, "peak_prefill_memory": null, "peak_decode_memory": 6567653464, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}28{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "b88d04948c6076d6b1e74562", "attempt_id": "38670f80572e", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:55.520316Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 598, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 168.28325006645173, "ttft_ms": 42.49120794702321, "decode_ms": 53.93745796754956}, "total_wall_ms": 266.03237504605204, "ttft_ms": 42.49120794702321, "input_usage": {"receiver_prompt_tokens": 598}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600507392, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6568931416, "mlx_cache_bytes": 4525622269, "process_rss_bytes": 6600507392}, "peak_prefill_memory": null, "peak_decode_memory": 6568931416, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}29{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "43812728b5c270b829d4f9ce", "attempt_id": "038b0bcc29f7", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:55.807395Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 606, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 179.05437492299825, "ttft_ms": 30.108542065136135, "decode_ms": 365.62775005586445}, "total_wall_ms": 576.2662909692153, "ttft_ms": 30.108542065136135, "input_usage": {"receiver_prompt_tokens": 606}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6600540160, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6571159640, "mlx_cache_bytes": 4894262269, "process_rss_bytes": 6600540160}, "peak_prefill_memory": null, "peak_decode_memory": 6571159640, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}30{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2bfef06d645d76d7f74dd897", "attempt_id": "1e7e8299b397", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:56.404836Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 604, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 197.55854201503098, "ttft_ms": 35.68808303680271, "decode_ms": 27.315416955389082}, "total_wall_ms": 262.4514999333769, "ttft_ms": 35.68808303680271, "input_usage": {"receiver_prompt_tokens": 604}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600572928, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6585266264, "mlx_cache_bytes": 5164942333, "process_rss_bytes": 6600572928}, "peak_prefill_memory": null, "peak_decode_memory": 6585266264, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}31{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "749a10f41570225b4bb45f60", "attempt_id": "e88897f5be11", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:56.690208Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 598, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 195.7303750095889, "ttft_ms": 37.23991697188467, "decode_ms": 33.89962494838983}, "total_wall_ms": 268.46675004344434, "ttft_ms": 37.23991697188467, "input_usage": {"receiver_prompt_tokens": 598}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600572928, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6568931416, "mlx_cache_bytes": 5164942333, "process_rss_bytes": 6600572928}, "peak_prefill_memory": null, "peak_decode_memory": 6568931416, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}32{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6ccb56e1b78581aa65d29afd", "attempt_id": "41e0321def09", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:56.982992Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1971}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 596, "message_tokens": 516, "message_utf8_bytes": 1775, "cache_tensor_bytes": null, "serialized_payload_bytes": 1775, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.8415419617667794, "decode_payload_ms": 0.0, "receiver_prefill_ms": 206.1636250000447, "ttft_ms": 52.30099998880178, "decode_ms": 22.620375035330653}, "total_wall_ms": 282.57141599897295, "ttft_ms": 52.30099998880178, "input_usage": {"receiver_prompt_tokens": 596}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600605696, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567358552, "mlx_cache_bytes": 5410014205, "process_rss_bytes": 6600605696}, "peak_prefill_memory": null, "peak_decode_memory": 6567358552, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "d2f5c011b1bc04cde05fd53e8eda8cce4f41dc708c77ba1e0b49bd3fb47d3958", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}33{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2d52bb8506077d32b22fdf34", "attempt_id": "2fdec005991b", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:57.291069Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 203, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 105.24599999189377, "ttft_ms": 44.75383390672505, "decode_ms": 24.731541983783245}, "total_wall_ms": 175.54037494119257, "ttft_ms": 44.75383390672505, "input_usage": {"receiver_prompt_tokens": 203}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600736768, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6345204848, "mlx_cache_bytes": 5567824893, "process_rss_bytes": 6600736768}, "peak_prefill_memory": null, "peak_decode_memory": 6345204848, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}34{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "960a53b6ffa03f96204e113b", "attempt_id": "c8f088125334", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:57.488109Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 209, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 59.651290997862816, "ttft_ms": 44.325374998152256, "decode_ms": 47.68295807298273}, "total_wall_ms": 152.65408309642226, "ttft_ms": 44.325374998152256, "input_usage": {"receiver_prompt_tokens": 209}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600769536, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6350185584, "mlx_cache_bytes": 5682119677, "process_rss_bytes": 6600769536}, "peak_prefill_memory": null, "peak_decode_memory": 6350185584, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}35{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "350f8498c2cc2bdba46117dc", "attempt_id": "f0eea4cdaefe", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:57.662250Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 209, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 92.06504200119525, "ttft_ms": 50.073124933987856, "decode_ms": 18.201000057160854}, "total_wall_ms": 161.36324999388307, "ttft_ms": 50.073124933987856, "input_usage": {"receiver_prompt_tokens": 209}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600769536, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6350185584, "mlx_cache_bytes": 5682119677, "process_rss_bytes": 6600769536}, "peak_prefill_memory": null, "peak_decode_memory": 6350185584, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}36{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "bed08623df2b233a51b643e8", "attempt_id": "019b3ff18573", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:57.845326Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 206, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 125.36499998532236, "ttft_ms": 43.34979096893221, "decode_ms": 50.19183293916285}, "total_wall_ms": 219.6721249492839, "ttft_ms": 43.34979096893221, "input_usage": {"receiver_prompt_tokens": 206}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600802304, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6357738656, "mlx_cache_bytes": 5805917181, "process_rss_bytes": 6600802304}, "peak_prefill_memory": null, "peak_decode_memory": 6357738656, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}37{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "da70f6f1d8274607021d1a65", "attempt_id": "aaf632e32046", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:58.086755Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 214, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 82.11695798672736, "ttft_ms": 47.32775001320988, "decode_ms": 48.785457969643176}, "total_wall_ms": 179.21474995091558, "ttft_ms": 47.32775001320988, "input_usage": {"receiver_prompt_tokens": 214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600818688, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6353364080, "mlx_cache_bytes": 5805917181, "process_rss_bytes": 6600818688}, "peak_prefill_memory": null, "peak_decode_memory": 6353364080, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}38{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5a548401bf8107c962e7f07f", "attempt_id": "fad0ca1f04b2", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:58.286423Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 212, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 94.01095903012902, "ttft_ms": 36.137750023044646, "decode_ms": 42.842958006076515}, "total_wall_ms": 173.7732500769198, "ttft_ms": 36.137750023044646, "input_usage": {"receiver_prompt_tokens": 212}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600835072, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6351889520, "mlx_cache_bytes": 5805917181, "process_rss_bytes": 6600835072}, "peak_prefill_memory": null, "peak_decode_memory": 6351889520, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}39{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "7d6d7ceed47419bf3def92f4", "attempt_id": "4f9c9ceb94ce", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:58.480151Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 206, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 82.23549998365343, "ttft_ms": 43.14154200255871, "decode_ms": 21.135332994163036}, "total_wall_ms": 147.38370792474598, "ttft_ms": 43.14154200255871, "input_usage": {"receiver_prompt_tokens": 206}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600851456, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6358295688, "mlx_cache_bytes": 5805917181, "process_rss_bytes": 6600851456}, "peak_prefill_memory": null, "peak_decode_memory": 6358295688, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}40{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "d649e2630c5f9f61268086dd", "attempt_id": "818276900582", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:58.649480Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 3}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 204, "message_tokens": 128, "message_utf8_bytes": 326, "cache_tensor_bytes": null, "serialized_payload_bytes": 326, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.7147079342976213, "decode_payload_ms": 0.0, "receiver_prefill_ms": 91.93212504032999, "ttft_ms": 19.05783300753683, "decode_ms": 29.006875003688037}, "total_wall_ms": 140.94204199500382, "ttft_ms": 19.05783300753683, "input_usage": {"receiver_prompt_tokens": 204}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600851456, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6345729136, "mlx_cache_bytes": 5917459453, "process_rss_bytes": 6600851456}, "peak_prefill_memory": null, "peak_decode_memory": 6345729136, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67dd5df4b6c2cc4b5034e9623e6e48ec476f1a5f3b53d4ed80fdb39c5a099d19", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}41{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "77ffefdf43d830d58ed924aa", "attempt_id": "99653bbd39bb", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:58.814598Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 583, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 155.91591701377183, "ttft_ms": 41.78220801986754, "decode_ms": 48.34500001743436}, "total_wall_ms": 247.34445801004767, "ttft_ms": 41.78220801986754, "input_usage": {"receiver_prompt_tokens": 583}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600949760, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6559592536, "mlx_cache_bytes": 6272041981, "process_rss_bytes": 6600949760}, "peak_prefill_memory": null, "peak_decode_memory": 6559592536, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}42{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "d526aabbe84d0dba86056194", "attempt_id": "c7be020467cc", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:59.084813Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 589, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 202.24424998741597, "ttft_ms": 51.58279195893556, "decode_ms": 46.79262498393655}, "total_wall_ms": 301.9260410219431, "ttft_ms": 51.58279195893556, "input_usage": {"receiver_prompt_tokens": 589}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601015296, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6563131480, "mlx_cache_bytes": 6615417853, "process_rss_bytes": 6601015296}, "peak_prefill_memory": null, "peak_decode_memory": 6563131480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}43{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2bebbc559d57148a86a9913b", "attempt_id": "bac7cc0fe14f", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:59.407181Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 589, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 148.44395802356303, "ttft_ms": 42.530416045337915, "decode_ms": 378.4955000737682}, "total_wall_ms": 570.7270840648562, "ttft_ms": 42.530416045337915, "input_usage": {"receiver_prompt_tokens": 589}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6601031680, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6563131480, "mlx_cache_bytes": 6615417853, "process_rss_bytes": 6601031680}, "peak_prefill_memory": null, "peak_decode_memory": 6563131480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,068,030", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "d1643b9a5cdbde943cffa17e0bd2ea3a63bd11939d2b80e5c53fe3f1d39a8e26"}44{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5ba4373fd5c1c6bf514fb3a7", "attempt_id": "b84710e9686f", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:49:59.998287Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 586, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 146.26754098571837, "ttft_ms": 21.77520899567753, "decode_ms": 183.84687509387732}, "total_wall_ms": 353.2489159842953, "ttft_ms": 21.77520899567753, "input_usage": {"receiver_prompt_tokens": 586}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 6601048064, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6561951832, "mlx_cache_bytes": 6856459261, "process_rss_bytes": 6601048064}, "peak_prefill_memory": null, "peak_decode_memory": 6561951832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}45{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "284d5c9a6da9e1176c093723", "attempt_id": "35eb8851efc9", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:00.374341Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 594, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 173.6891670152545, "ttft_ms": 43.01970906089991, "decode_ms": 325.49429091159254}, "total_wall_ms": 543.6962500680238, "ttft_ms": 43.01970906089991, "input_usage": {"receiver_prompt_tokens": 594}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6601080832, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6566572120, "mlx_cache_bytes": 7013155837, "process_rss_bytes": 6601080832}, "peak_prefill_memory": null, "peak_decode_memory": 6566572120, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}46{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "745fe7f5792c4ff45fbe1259", "attempt_id": "f0fef70ba26a", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:00.939653Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 592, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 177.32991697266698, "ttft_ms": 51.515334052965045, "decode_ms": 48.80475008394569}, "total_wall_ms": 278.98929093498737, "ttft_ms": 51.515334052965045, "input_usage": {"receiver_prompt_tokens": 592}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600949760, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6585167960, "mlx_cache_bytes": 7278478333, "process_rss_bytes": 6600949760}, "peak_prefill_memory": null, "peak_decode_memory": 6585167960, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}47{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "1b5f86671906a5ce0e014cdc", "attempt_id": "b5e707987966", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:01.239542Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 586, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 201.5844159759581, "ttft_ms": 31.360000022687018, "decode_ms": 23.63358298316598}, "total_wall_ms": 257.8552080085501, "ttft_ms": 31.360000022687018, "input_usage": {"receiver_prompt_tokens": 586}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600982528, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6561951832, "mlx_cache_bytes": 7278478333, "process_rss_bytes": 6600982528}, "peak_prefill_memory": null, "peak_decode_memory": 6561951832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}48{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f4753707bacf518cec2c7e8b", "attempt_id": "ab1f2bd44b6c", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:01.517818Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 14}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 584, "message_tokens": 508, "message_utf8_bytes": 1231, "cache_tensor_bytes": null, "serialized_payload_bytes": 1231, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.127957995980978, "decode_payload_ms": 0.0, "receiver_prefill_ms": 190.75366598553956, "ttft_ms": 39.51524989679456, "decode_ms": 43.95299998577684}, "total_wall_ms": 275.4939589649439, "ttft_ms": 39.51524989679456, "input_usage": {"receiver_prompt_tokens": 584}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600998912, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6559985752, "mlx_cache_bytes": 7518635005, "process_rss_bytes": 6600998912}, "peak_prefill_memory": null, "peak_decode_memory": 6559985752, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "386755a6041624133cae3e906274e1dbaa1ac732378d43f385e9581f5c9aba44", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}49{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ff503f4991b69d71a9cbbcd9", "attempt_id": "cd5a0a253b31", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:01.815582Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 512, "n_sentences": 24}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 587, "message_tokens": 512, "message_utf8_bytes": 1358, "cache_tensor_bytes": null, "serialized_payload_bytes": 1358, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.1087090242654085, "decode_payload_ms": 0.0, "receiver_prefill_ms": 182.11904203053564, "ttft_ms": 42.147999978624284, "decode_ms": 335.11224994435906}, "total_wall_ms": 561.8233749410138, "ttft_ms": 42.147999978624284, "input_usage": {"receiver_prompt_tokens": 587}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6600835072, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6562246744, "mlx_cache_bytes": 7759971325, "process_rss_bytes": 6600835072}, "peak_prefill_memory": null, "peak_decode_memory": 6562246744, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "9c6f1d7683c25b99905915cd50af46f8624d0e046d60064afa1f8536d2053414", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}50{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "cfed9f490577aa19cffef409", "attempt_id": "aad1d7b7f79b", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:02.398863Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 25}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 589, "message_tokens": 508, "message_utf8_bytes": 1477, "cache_tensor_bytes": null, "serialized_payload_bytes": 1477, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.0590419406071305, "decode_payload_ms": 0.0, "receiver_prefill_ms": 175.15408399049193, "ttft_ms": 46.27291695214808, "decode_ms": 337.7104999963194}, "total_wall_ms": 561.484833015129, "ttft_ms": 46.27291695214808, "input_usage": {"receiver_prompt_tokens": 589}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6600835072, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6563131480, "mlx_cache_bytes": 7759971325, "process_rss_bytes": 6600835072}, "peak_prefill_memory": null, "peak_decode_memory": 6563131480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0c282c4bd4288101fb9f0b7c3f7960301f9b9cae0e1b006f3d55fa33cb6c3bdc", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}51{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "e7d124b13680912c2a334bab", "attempt_id": "561ac6b167af", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:02.981197Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 499, "n_sentences": 24}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 580, "message_tokens": 499, "message_utf8_bytes": 1320, "cache_tensor_bytes": null, "serialized_payload_bytes": 1320, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.0496670147404075, "decode_payload_ms": 0.0, "receiver_prefill_ms": 167.33445797581226, "ttft_ms": 38.87175000272691, "decode_ms": 289.80229201260954}, "total_wall_ms": 498.2860420132056, "ttft_ms": 38.87175000272691, "input_usage": {"receiver_prompt_tokens": 580}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6600835072, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6577401944, "mlx_cache_bytes": 8019936253, "process_rss_bytes": 6600835072}, "peak_prefill_memory": null, "peak_decode_memory": 6577401944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "77c2c83f97318ad674e337c84e8706903022a99e330a15111b84b24346998e7c", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}52{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "27baf0f22614de8b3768ca95", "attempt_id": "9e8d3021295e", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:03.500827Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 499, "n_sentences": 17}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 577, "message_tokens": 499, "message_utf8_bytes": 1439, "cache_tensor_bytes": null, "serialized_payload_bytes": 1439, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.9218748891726136, "decode_payload_ms": 0.0, "receiver_prefill_ms": 148.2660420006141, "ttft_ms": 52.42104199714959, "decode_ms": 207.91524997912347}, "total_wall_ms": 410.8636670280248, "ttft_ms": 52.42104199714959, "input_usage": {"receiver_prompt_tokens": 577}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 6600949760, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6555218008, "mlx_cache_bytes": 8334902269, "process_rss_bytes": 6600949760}, "peak_prefill_memory": null, "peak_decode_memory": 6555218008, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "8c78c2c57027b9340eda90e83f732ddd57af36c5177bd2241b5f21f5452cfb30", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}53{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2fe05c9027e879b87d76e4cd", "attempt_id": "834114cbda04", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:03.932833Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 511, "n_sentences": 14}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 597, "message_tokens": 511, "message_utf8_bytes": 1212, "cache_tensor_bytes": null, "serialized_payload_bytes": 1212, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.136208069510758, "decode_payload_ms": 0.0, "receiver_prefill_ms": 209.54304095357656, "ttft_ms": 48.278250033035874, "decode_ms": 325.0983749749139}, "total_wall_ms": 585.3710420196876, "ttft_ms": 48.278250033035874, "input_usage": {"receiver_prompt_tokens": 597}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6600949760, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567653464, "mlx_cache_bytes": 8580269053, "process_rss_bytes": 6600949760}, "peak_prefill_memory": null, "peak_decode_memory": 6567653464, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "b1b71eb7de8b1ef4eb5d401effd828bab160a4a00b6b0c9c62db233a9dc6cdf7", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}54{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "bc2675868d450a10dfb9c39e", "attempt_id": "f70e965119b1", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:04.541246Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 507, "n_sentences": 26}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 591, "message_tokens": 507, "message_utf8_bytes": 1431, "cache_tensor_bytes": null, "serialized_payload_bytes": 1431, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.1420410592108965, "decode_payload_ms": 0.0, "receiver_prefill_ms": 189.75595792289823, "ttft_ms": 43.90158294700086, "decode_ms": 364.7835840238258}, "total_wall_ms": 601.0728749679402, "ttft_ms": 43.90158294700086, "input_usage": {"receiver_prompt_tokens": 591}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6600966144, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6582136920, "mlx_cache_bytes": 8823178237, "process_rss_bytes": 6600966144}, "peak_prefill_memory": null, "peak_decode_memory": 6582136920, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "SP-38G-4217", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e7c991ceabb14f3dd74de0b7f83992695d4c4b66708e36722b855d130f504f94", "output_hash": "e643c21a78559b96fd65325d9ce779b7b8fd6a47c53cf7c2cda1efa8412458c4"}55{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f70e52e5a1e040bfbf73181e", "attempt_id": "bee678b24cf4", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:05.164538Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 509, "n_sentences": 22}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 587, "message_tokens": 509, "message_utf8_bytes": 1514, "cache_tensor_bytes": null, "serialized_payload_bytes": 1514, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.2312920298427343, "decode_payload_ms": 0.0, "receiver_prefill_ms": 195.89691690634936, "ttft_ms": 37.64795802999288, "decode_ms": 43.63908397499472}, "total_wall_ms": 279.62329098954797, "ttft_ms": 37.64795802999288, "input_usage": {"receiver_prompt_tokens": 587}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600982528, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6562246744, "mlx_cache_bytes": 8823178237, "process_rss_bytes": 6600982528}, "peak_prefill_memory": null, "peak_decode_memory": 6562246744, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "b968cdd6fcc361d0982c0da716336ce5ade8e57a2374e310262166750a963a0d", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}56{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6b2c79c4c442a62acfc8e721", "attempt_id": "884ca4710385", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:05.465061Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 510, "n_sentences": 22}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 586, "message_tokens": 510, "message_utf8_bytes": 1513, "cache_tensor_bytes": null, "serialized_payload_bytes": 1513, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.013041939586401, "decode_payload_ms": 0.0, "receiver_prefill_ms": 189.1854169080034, "ttft_ms": 18.651874968782067, "decode_ms": 26.465042028576136}, "total_wall_ms": 236.82258394546807, "ttft_ms": 18.651874968782067, "input_usage": {"receiver_prompt_tokens": 586}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6600982528, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6561951832, "mlx_cache_bytes": 8823178237, "process_rss_bytes": 6600982528}, "peak_prefill_memory": null, "peak_decode_memory": 6561951832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f6abba20464ca36454f355f0746779b994274921acba0d41c40660a00845ed1f", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}57{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "bd50c544969cd0b73594684f", "attempt_id": "d1525a6dc027", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:11.080862Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 207, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 86.47970808669925, "ttft_ms": 34.69141595996916, "decode_ms": 33.22983405087143}, "total_wall_ms": 155.22262500599027, "ttft_ms": 34.69141595996916, "input_usage": {"receiver_prompt_tokens": 207, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601244672, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6358688904, "mlx_cache_bytes": 8943502333, "process_rss_bytes": 6601244672}, "peak_prefill_memory": null, "peak_decode_memory": 6358688904, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}58{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "001c12267a45fe00a59db1be", "attempt_id": "9d31cda253e9", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:11.259070Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 213, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 80.06029098760337, "ttft_ms": 46.73562501557171, "decode_ms": 37.23804093897343}, "total_wall_ms": 164.89437501877546, "ttft_ms": 46.73562501557171, "input_usage": {"receiver_prompt_tokens": 213, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601244672, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6351725680, "mlx_cache_bytes": 9059894269, "process_rss_bytes": 6601244672}, "peak_prefill_memory": null, "peak_decode_memory": 6351725680, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}59{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "807e4cc7994c09447b603950", "attempt_id": "a9ad279d120b", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:11.443689Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 213, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 88.87387509457767, "ttft_ms": 46.50487506296486, "decode_ms": 22.77625002898276}, "total_wall_ms": 158.9888749876991, "ttft_ms": 46.50487506296486, "input_usage": {"receiver_prompt_tokens": 213, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601244672, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6351725680, "mlx_cache_bytes": 9059894269, "process_rss_bytes": 6601244672}, "peak_prefill_memory": null, "peak_decode_memory": 6351725680, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}60{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2d5b038eadcfcb192be47771", "attempt_id": "f7131dbeb50b", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:11.623283Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 210, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 79.21091699972749, "ttft_ms": 43.99662499781698, "decode_ms": 42.46820800472051}, "total_wall_ms": 166.74104100093246, "ttft_ms": 43.99662499781698, "input_usage": {"receiver_prompt_tokens": 210, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601261056, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6350840944, "mlx_cache_bytes": 9133556733, "process_rss_bytes": 6601261056}, "peak_prefill_memory": null, "peak_decode_memory": 6350840944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}61{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "4e587557a9581e25d551c063", "attempt_id": "f682d9d14e23", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:11.811245Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 218, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 87.74291700683534, "ttft_ms": 33.67295803036541, "decode_ms": 356.4188750460744}, "total_wall_ms": 478.7968749878928, "ttft_ms": 33.67295803036541, "input_usage": {"receiver_prompt_tokens": 218, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6601277440, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6367814792, "mlx_cache_bytes": 9252832253, "process_rss_bytes": 6601277440}, "peak_prefill_memory": null, "peak_decode_memory": 6367814792, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}62{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6ce8dab9103c1c633ee0ad19", "attempt_id": "1b0314a4665c", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:12.311221Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 216, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 83.17016693763435, "ttft_ms": 56.20595801156014, "decode_ms": 31.98158403392881}, "total_wall_ms": 172.2495419671759, "ttft_ms": 56.20595801156014, "input_usage": {"receiver_prompt_tokens": 216, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601293824, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6354412656, "mlx_cache_bytes": 9328591869, "process_rss_bytes": 6601293824}, "peak_prefill_memory": null, "peak_decode_memory": 6354412656, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}63{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "03f25a96c38315f546ef3f7a", "attempt_id": "8426d55df1d3", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:12.505465Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 210, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 94.84495909418911, "ttft_ms": 33.99795794393867, "decode_ms": 38.60362502746284}, "total_wall_ms": 168.34004106931388, "ttft_ms": 33.99795794393867, "input_usage": {"receiver_prompt_tokens": 210, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601293824, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6350840944, "mlx_cache_bytes": 9328591869, "process_rss_bytes": 6601293824}, "peak_prefill_memory": null, "peak_decode_memory": 6350840944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}64{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "d87ff6d5b0d9bcceffbffbbc", "attempt_id": "076dcc0c99b3", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:12.695828Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 208, "message_tokens": 128, "message_utf8_bytes": 288, "cache_tensor_bytes": null, "serialized_payload_bytes": 288, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5357.089624973014, "decode_payload_ms": 0.0, "receiver_prefill_ms": 84.99754196964204, "ttft_ms": 19.704332924447954, "decode_ms": 37.26479201577604}, "total_wall_ms": 142.82641594763845, "ttft_ms": 19.704332924447954, "input_usage": {"receiver_prompt_tokens": 208, "sender_input_tokens": 2086, "sender_output_tokens": 128, "sender_prefill_ms": 572.2841250244528, "sender_ttft_ms": 43.00425003748387, "sender_decode_ms": 4739.774457993917}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601293824, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6359868552, "mlx_cache_bytes": 9328591869, "process_rss_bytes": 6601293824}, "peak_prefill_memory": null, "peak_decode_memory": 6359868552, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c1f5c1b08a795ed7110a28f89139d5d4e36268379c8be60bcb7449d0cf723f13", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}65{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "60db059748d0a3b05967b502", "attempt_id": "828ee0224cea", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:33.128851Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 591, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 203.14004202373326, "ttft_ms": 28.74737500678748, "decode_ms": 394.19945899862796}, "total_wall_ms": 627.724124933593, "ttft_ms": 28.74737500678748, "input_usage": {"receiver_prompt_tokens": 591, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6601375744, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6582136920, "mlx_cache_bytes": 9500820477, "process_rss_bytes": 6601375744}, "peak_prefill_memory": null, "peak_decode_memory": 6582136920, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}66{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "b0b0c81f969c1736283bcfe7", "attempt_id": "9bb1b19b0b3d", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:33.783215Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 597, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 183.51595802232623, "ttft_ms": 19.946582964621484, "decode_ms": 41.30941699258983}, "total_wall_ms": 246.46916706115007, "ttft_ms": 19.946582964621484, "input_usage": {"receiver_prompt_tokens": 597, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601375744, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567653464, "mlx_cache_bytes": 9500820477, "process_rss_bytes": 6601375744}, "peak_prefill_memory": null, "peak_decode_memory": 6567653464, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}67{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "99de6273c9600fddb438f1d7", "attempt_id": "2c4fe8796cf3", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:34.054357Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 597, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 232.18620801344514, "ttft_ms": 30.138583038933575, "decode_ms": 33.883374999277294}, "total_wall_ms": 297.7294169832021, "ttft_ms": 30.138583038933575, "input_usage": {"receiver_prompt_tokens": 597, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601375744, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567653464, "mlx_cache_bytes": 9500820477, "process_rss_bytes": 6601375744}, "peak_prefill_memory": null, "peak_decode_memory": 6567653464, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}68{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "972b2521690d285b8bc67c21", "attempt_id": "df2e83a51555", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:34.374794Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 594, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 224.3669159943238, "ttft_ms": 22.97604107297957, "decode_ms": 17.95237499754876}, "total_wall_ms": 266.74049999564886, "ttft_ms": 22.97604107297957, "input_usage": {"receiver_prompt_tokens": 594, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601375744, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6566572120, "mlx_cache_bytes": 9500820477, "process_rss_bytes": 6601375744}, "peak_prefill_memory": null, "peak_decode_memory": 6566572120, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}69{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "00febe37781016a142d26964", "attempt_id": "65ea1c5d1d91", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:34.664475Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 602, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 207.15975004713982, "ttft_ms": 46.175166964530945, "decode_ms": 368.2772920001298}, "total_wall_ms": 623.0741670588031, "ttft_ms": 46.175166964530945, "input_usage": {"receiver_prompt_tokens": 602, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6601408512, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6581891160, "mlx_cache_bytes": 9748349949, "process_rss_bytes": 6601408512}, "peak_prefill_memory": null, "peak_decode_memory": 6581891160, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}70{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "85de3682876a5ad12924390a", "attempt_id": "2de5beb86693", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:35.311618Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 600, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 165.15562497079372, "ttft_ms": 48.18745795637369, "decode_ms": 388.31354200374335}, "total_wall_ms": 603.0747499316931, "ttft_ms": 48.18745795637369, "input_usage": {"receiver_prompt_tokens": 600, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6601408512, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567260248, "mlx_cache_bytes": 9906619389, "process_rss_bytes": 6601408512}, "peak_prefill_memory": null, "peak_decode_memory": 6567260248, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "SP-38G-4217", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "e643c21a78559b96fd65325d9ce779b7b8fd6a47c53cf7c2cda1efa8412458c4"}71{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "040faea8dc90b744a248bfbb", "attempt_id": "89c386e78ead", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:35.937650Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 594, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 150.53695801179856, "ttft_ms": 26.867000036872923, "decode_ms": 31.897875014692545}, "total_wall_ms": 210.681667085737, "ttft_ms": 26.867000036872923, "input_usage": {"receiver_prompt_tokens": 594, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601408512, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6566572120, "mlx_cache_bytes": 9906619389, "process_rss_bytes": 6601408512}, "peak_prefill_memory": null, "peak_decode_memory": 6566572120, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}72{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9a38432c6d073693cc7abcef", "attempt_id": "d3eb20e34a83", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:36.170131Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 592, "message_tokens": 512, "message_utf8_bytes": 1196, "cache_tensor_bytes": null, "serialized_payload_bytes": 1196, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20267.59104197845, "decode_payload_ms": 0.0, "receiver_prefill_ms": 193.10900010168552, "ttft_ms": 48.168833018280566, "decode_ms": 30.483458074741066}, "total_wall_ms": 273.1704580364749, "ttft_ms": 48.168833018280566, "input_usage": {"receiver_prompt_tokens": 592, "sender_input_tokens": 2087, "sender_output_tokens": 512, "sender_prefill_ms": 548.8623749697581, "sender_ttft_ms": 49.25675003323704, "sender_decode_ms": 19667.46804106515}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6601408512, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6582726744, "mlx_cache_bytes": 9906619389, "process_rss_bytes": 6601408512}, "peak_prefill_memory": null, "peak_decode_memory": 6582726744, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f811e3d2b4b96fb92f1cc9651f50294bf8a18ad7c33cafac648e5969f236fb1e", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}73{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c5d80addd1f25a970ef70a60", "attempt_id": "dc60effed0dc", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:36.467592Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 121, "message_tokens": 43, "message_utf8_bytes": 168, "cache_tensor_bytes": null, "serialized_payload_bytes": 168, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2326.683791936375, "decode_payload_ms": 0.0, "receiver_prefill_ms": 55.99925003480166, "ttft_ms": 30.39112500846386, "decode_ms": 423.52000006940216}, "total_wall_ms": 2837.184499949217, "ttft_ms": 30.39112500846386, "input_usage": {"receiver_prompt_tokens": 121, "sender_input_tokens": 2082, "sender_output_tokens": 43, "sender_prefill_ms": 575.4174999892712, "sender_ttft_ms": 41.759541956707835, "sender_decode_ms": 1707.3157080449164}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6558646272, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6278603376, "mlx_cache_bytes": 10015048701, "process_rss_bytes": 6558646272}, "peak_prefill_memory": null, "peak_decode_memory": 6278603376, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "67918d4922cc4a215f8af73fc75418b9db0043e809df8ec4e58e84d6f417633c", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}74{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "059c1e317cf8f7f98d662f51", "attempt_id": "f29c392371ba", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:39.328730Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 170, "message_tokens": 86, "message_utf8_bytes": 269, "cache_tensor_bytes": null, "serialized_payload_bytes": 269, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3948.0282498989254, "decode_payload_ms": 0.0, "receiver_prefill_ms": 85.46241594012827, "ttft_ms": 47.27683402597904, "decode_ms": 281.0025420039892}, "total_wall_ms": 4362.543457886204, "ttft_ms": 47.27683402597904, "input_usage": {"receiver_prompt_tokens": 170, "sender_input_tokens": 2088, "sender_output_tokens": 86, "sender_prefill_ms": 572.1418749308214, "sender_ttft_ms": 46.04549997020513, "sender_decode_ms": 3327.9037920292467}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6558760960, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6319449200, "mlx_cache_bytes": 10154214397, "process_rss_bytes": 6558760960}, "peak_prefill_memory": null, "peak_decode_memory": 6319449200, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "13f8fdf17734f24bbe35b808c94ee3e08ab82732e7aec7db49c33a3a0353ffc4", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}75{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "28463ed003820e507349b8ea", "attempt_id": "fe605ec1b8e3", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:43.712225Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 120, "message_tokens": 36, "message_utf8_bytes": 120, "cache_tensor_bytes": null, "serialized_payload_bytes": 120, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2028.6041660001501, "decode_payload_ms": 0.0, "receiver_prefill_ms": 65.41991699486971, "ttft_ms": 52.16733308043331, "decode_ms": 280.0687920534983}, "total_wall_ms": 2426.891708979383, "ttft_ms": 52.16733308043331, "input_usage": {"receiver_prompt_tokens": 120, "sender_input_tokens": 2088, "sender_output_tokens": 36, "sender_prefill_ms": 611.6607079748064, "sender_ttft_ms": 39.29641598369926, "sender_decode_ms": 1375.6340839900076}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6558777344, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6278079088, "mlx_cache_bytes": 10196157437, "process_rss_bytes": 6558777344}, "peak_prefill_memory": null, "peak_decode_memory": 6278079088, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "69f0de825fd0240070d8315da99518a3de438de80cf7ebf353b4ba57798fb79d", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}76{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "cae7610011d07e2d4ed7f376", "attempt_id": "d7ccedc639a2", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:46.160460Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 155, "message_tokens": 74, "message_utf8_bytes": 236, "cache_tensor_bytes": null, "serialized_payload_bytes": 236, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3465.2858750196174, "decode_payload_ms": 0.0, "receiver_prefill_ms": 108.12795802485198, "ttft_ms": 44.621291919611394, "decode_ms": 108.44966606236994}, "total_wall_ms": 3727.6338329538703, "ttft_ms": 44.621291919611394, "input_usage": {"receiver_prompt_tokens": 155, "sender_input_tokens": 2085, "sender_output_tokens": 74, "sender_prefill_ms": 554.2692090384662, "sender_ttft_ms": 41.221374995075166, "sender_decode_ms": 2867.85387492273}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 6557892608, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6306620528, "mlx_cache_bytes": 10334454781, "process_rss_bytes": 6557892608}, "peak_prefill_memory": null, "peak_decode_memory": 6306620528, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "6870ea9f2332f5028ea3bd157a4cd79185e0c56d7dbdff6c6d165d96ab97ec94", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}77{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "e52c34ec69177456ba83f79b", "attempt_id": "b2394cae2d4a", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:49.909430Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 251, "message_tokens": 162, "message_utf8_bytes": 521, "cache_tensor_bytes": null, "serialized_payload_bytes": 521, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6819.481833954342, "decode_payload_ms": 0.0, "receiver_prefill_ms": 76.50120905600488, "ttft_ms": 42.372124968096614, "decode_ms": 302.0131249213591}, "total_wall_ms": 7241.7372920317575, "ttft_ms": 42.372124968096614, "input_usage": {"receiver_prompt_tokens": 251, "sender_input_tokens": 2093, "sender_output_tokens": 162, "sender_prefill_ms": 707.0347499102354, "sender_ttft_ms": 52.32829204760492, "sender_decode_ms": 6057.84904200118}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6388383744, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6397617264, "mlx_cache_bytes": 10505012221, "process_rss_bytes": 6388383744}, "peak_prefill_memory": null, "peak_decode_memory": 6397617264, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3463b753dd8829ea6fec059b9cd58dcc1a7c02bf15c4fad6b7612262255b579b", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}78{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8bcead4a1aaf963ae9e946bc", "attempt_id": "00e556a7cd42", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:50:57.176355Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 147, "message_tokens": 60, "message_utf8_bytes": 240, "cache_tensor_bytes": null, "serialized_payload_bytes": 240, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2926.3977089431137, "decode_payload_ms": 0.0, "receiver_prefill_ms": 41.571542038582265, "ttft_ms": 28.354624984785914, "decode_ms": 413.99049991741776}, "total_wall_ms": 3411.1245409585536, "ttft_ms": 28.354624984785914, "input_usage": {"receiver_prompt_tokens": 147, "sender_input_tokens": 2091, "sender_output_tokens": 60, "sender_prefill_ms": 583.3549579838291, "sender_ttft_ms": 48.79212495870888, "sender_decode_ms": 2292.3897500149906}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6388662272, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6300279920, "mlx_cache_bytes": 10636116989, "process_rss_bytes": 6388662272}, "peak_prefill_memory": null, "peak_decode_memory": 6300279920, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "SP-38G-4217", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "bf25cc6aed1c6ade7cbaeeb4a635af6e9dec714e42a0a6e9bea2ffd1887bdac9", "output_hash": "e643c21a78559b96fd65325d9ce779b7b8fd6a47c53cf7c2cda1efa8412458c4"}79{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ef0ae0c957bbe5f7f9be4455", "attempt_id": "9a7a6d5e58cf", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:00.607423Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 242, "message_tokens": 161, "message_utf8_bytes": 428, "cache_tensor_bytes": null, "serialized_payload_bytes": 428, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6918.077458045445, "decode_payload_ms": 0.0, "receiver_prefill_ms": 84.76508292369545, "ttft_ms": 42.02079097740352, "decode_ms": 36.63241700269282}, "total_wall_ms": 7082.401582971215, "ttft_ms": 42.02079097740352, "input_usage": {"receiver_prompt_tokens": 242, "sender_input_tokens": 2085, "sender_output_tokens": 161, "sender_prefill_ms": 644.1854999866337, "sender_ttft_ms": 47.369541018269956, "sender_decode_ms": 6224.478416959755}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6389055488, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6375466096, "mlx_cache_bytes": 10782000125, "process_rss_bytes": 6389055488}, "peak_prefill_memory": null, "peak_decode_memory": 6375466096, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4fd85a20e1dae1cfd12c1496aa35c7f6b9f63ec4f9e3954d6e8325f8a4da5f4d", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}80{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0ffcfc752a12e13bd6f61512", "attempt_id": "945fdb624ecd", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:07.708882Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 181, "message_tokens": 102, "message_utf8_bytes": 340, "cache_tensor_bytes": null, "serialized_payload_bytes": 340, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4510.75499993749, "decode_payload_ms": 0.0, "receiver_prefill_ms": 97.02654206193984, "ttft_ms": 33.74350001104176, "decode_ms": 34.6085000783205}, "total_wall_ms": 4677.098875050433, "ttft_ms": 33.74350001104176, "input_usage": {"receiver_prompt_tokens": 181, "sender_input_tokens": 2083, "sender_output_tokens": 102, "sender_prefill_ms": 567.8197500528768, "sender_ttft_ms": 36.7324169492349, "sender_decode_ms": 3904.24854203593}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6388269056, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6327739504, "mlx_cache_bytes": 10909139965, "process_rss_bytes": 6388269056}, "peak_prefill_memory": null, "peak_decode_memory": 6327739504, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3b75210dd375ba950bf4b73febd17778e61945603a415dc22d9590c42b1773b0", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}81{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8701c581dbc14f5227d24a15", "attempt_id": "3f8040a810ad", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:18.234519Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 137, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 152.1300420863554, "ttft_ms": 66.38229195959866, "decode_ms": 65.39833301212639}, "total_wall_ms": 285.24745802860707, "ttft_ms": 66.38229195959866, "input_usage": {"receiver_prompt_tokens": 137, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284525568, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6291530864, "mlx_cache_bytes": 11015046141, "process_rss_bytes": 6284525568}, "peak_prefill_memory": null, "peak_decode_memory": 6291530864, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}82{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "4332bd474814fd60a1da05ed", "attempt_id": "366bf77ee951", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:18.543040Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 143, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 138.74879200011492, "ttft_ms": 24.923749966546893, "decode_ms": 23.907874943688512}, "total_wall_ms": 188.33429098594934, "ttft_ms": 24.923749966546893, "input_usage": {"receiver_prompt_tokens": 143, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284787712, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6304130160, "mlx_cache_bytes": 11100111869, "process_rss_bytes": 6284787712}, "peak_prefill_memory": null, "peak_decode_memory": 6304130160, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}83{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "17730f009661178d0635a39e", "attempt_id": "11452778265a", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:18.752680Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 143, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 82.52950001042336, "ttft_ms": 42.7512499736622, "decode_ms": 46.38404201250523}, "total_wall_ms": 172.70062502939254, "ttft_ms": 42.7512499736622, "input_usage": {"receiver_prompt_tokens": 143, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284836864, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6303622280, "mlx_cache_bytes": 11100701693, "process_rss_bytes": 6284836864}, "peak_prefill_memory": null, "peak_decode_memory": 6303622280, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}84{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6d4124f6a346abb990d43447", "attempt_id": "8ad9e20eb066", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:18.947072Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 140, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 85.99408308509737, "ttft_ms": 49.946665996685624, "decode_ms": 44.71358307637274}, "total_wall_ms": 181.53095804154873, "ttft_ms": 49.946665996685624, "input_usage": {"receiver_prompt_tokens": 140, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284902400, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6296282224, "mlx_cache_bytes": 11202921469, "process_rss_bytes": 6284902400}, "peak_prefill_memory": null, "peak_decode_memory": 6296282224, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}85{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a81983359faa7e0580c80157", "attempt_id": "7f2d9dc14deb", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:19.149874Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 148, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 141.34308299981058, "ttft_ms": 52.11175000295043, "decode_ms": 54.08625002019107}, "total_wall_ms": 248.3807080425322, "ttft_ms": 52.11175000295043, "input_usage": {"receiver_prompt_tokens": 148, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284967936, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6300804208, "mlx_cache_bytes": 11283792893, "process_rss_bytes": 6284967936}, "peak_prefill_memory": null, "peak_decode_memory": 6300804208, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}86{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "49d598190933f0afabf5e21c", "attempt_id": "0a4884b5bd7d", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:19.419645Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 146, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 88.31079106312245, "ttft_ms": 65.96333300694823, "decode_ms": 56.599291041493416}, "total_wall_ms": 211.74404106568545, "ttft_ms": 65.96333300694823, "input_usage": {"receiver_prompt_tokens": 146, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6285066240, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6299755632, "mlx_cache_bytes": 11334910973, "process_rss_bytes": 6285066240}, "peak_prefill_memory": null, "peak_decode_memory": 6299755632, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}87{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "114d7f2656eacf1528ebd27f", "attempt_id": "8b3e5c0a31e2", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:19.650713Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 140, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 78.88616598211229, "ttft_ms": 37.94833295978606, "decode_ms": 18.885375000536442}, "total_wall_ms": 136.7663339478895, "ttft_ms": 37.94833295978606, "input_usage": {"receiver_prompt_tokens": 140, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6285131776, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6295676016, "mlx_cache_bytes": 11335484413, "process_rss_bytes": 6285131776}, "peak_prefill_memory": null, "peak_decode_memory": 6295676016, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}88{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3396ea923653f571f36e80f0", "attempt_id": "5d35ccda8dd3", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:19.810367Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 138, "message_tokens": 47, "message_utf8_bytes": 113, "cache_tensor_bytes": null, "serialized_payload_bytes": 113, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5828.520458075218, "decode_payload_ms": 0.0, "receiver_prefill_ms": 107.40645905025303, "ttft_ms": 77.94937503058463, "decode_ms": 65.68274996243417}, "total_wall_ms": 251.8139169551432, "ttft_ms": 77.94937503058463, "input_usage": {"receiver_prompt_tokens": 138, "sender_input_tokens": 2115, "sender_output_tokens": 128, "sender_prefill_ms": 753.3047500764951, "sender_ttft_ms": 46.19274998549372, "sender_decode_ms": 5025.524459080771}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6285164544, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6294889584, "mlx_cache_bytes": 11383981053, "process_rss_bytes": 6285164544}, "peak_prefill_memory": null, "peak_decode_memory": 6294889584, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e2e9d8372d61bc2abf4d92635ab8105eb20a4de648105008806bc2d8f7c4d96", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}89{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "65223841c0da0d6fd8422676", "attempt_id": "02a458c506d4", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:50.076247Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 434, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 364.8830000311136, "ttft_ms": 33.16437499597669, "decode_ms": 43.701917049475014}, "total_wall_ms": 443.91549995634705, "ttft_ms": 33.16437499597669, "input_usage": {"receiver_prompt_tokens": 434, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284083200, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6536111924, "mlx_cache_bytes": 11729046209, "process_rss_bytes": 6284083200}, "peak_prefill_memory": null, "peak_decode_memory": 6536111924, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}90{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "7a9127107893ed6aae4a4e6e", "attempt_id": "dfb72412f5aa", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:50.547946Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 440, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 265.50949993543327, "ttft_ms": 32.37508295569569, "decode_ms": 54.065417032688856}, "total_wall_ms": 353.77845796756446, "ttft_ms": 32.37508295569569, "input_usage": {"receiver_prompt_tokens": 440, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284197888, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6531884864, "mlx_cache_bytes": 12040393117, "process_rss_bytes": 6284197888}, "peak_prefill_memory": null, "peak_decode_memory": 6531884864, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}91{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "65f8820ba46bd4578a6d92b4", "attempt_id": "b57e18ac7621", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:50.924594Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 440, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 245.53833401296288, "ttft_ms": 56.499166996218264, "decode_ms": 47.08337504416704}, "total_wall_ms": 350.54616699926555, "ttft_ms": 56.499166996218264, "input_usage": {"receiver_prompt_tokens": 440, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284263424, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6538536816, "mlx_cache_bytes": 12062003613, "process_rss_bytes": 6284263424}, "peak_prefill_memory": null, "peak_decode_memory": 6538536816, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}92{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8442e0bb92aef48fd05ac2ce", "attempt_id": "b46d6970a724", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:51.299045Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 437, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 277.85641700029373, "ttft_ms": 42.986916028894484, "decode_ms": 56.898207985796034}, "total_wall_ms": 379.1518750367686, "ttft_ms": 42.986916028894484, "input_usage": {"receiver_prompt_tokens": 437, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6284279808, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6537684812, "mlx_cache_bytes": 12301341085, "process_rss_bytes": 6284279808}, "peak_prefill_memory": null, "peak_decode_memory": 6537684812, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}93{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "43942fc2671f1164f4a366a8", "attempt_id": "2b7fe200dd67", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:51.701094Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 445, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 323.8183750072494, "ttft_ms": 50.72587507311255, "decode_ms": 575.3746660193428}, "total_wall_ms": 951.2672499986365, "ttft_ms": 50.72587507311255, "input_usage": {"receiver_prompt_tokens": 445, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6280593408, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6543992672, "mlx_cache_bytes": 12632430733, "process_rss_bytes": 6280593408}, "peak_prefill_memory": null, "peak_decode_memory": 6543992672, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172 NOK", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.6666666666666666}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "ce5dae2b035af8d38bda9e7d1f1cb7eb7e4141e20298d86d8bce5fe5ac6c0246"}94{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "bc08d3f07374b4d8666c8240", "attempt_id": "70a270ecc729", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:52.680388Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 443, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 221.60295792855322, "ttft_ms": 59.783500037156045, "decode_ms": 472.54674998112023}, "total_wall_ms": 755.125334020704, "ttft_ms": 59.783500037156045, "input_usage": {"receiver_prompt_tokens": 443, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6279692288, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6542813024, "mlx_cache_bytes": 12875045005, "process_rss_bytes": 6279692288}, "peak_prefill_memory": null, "peak_decode_memory": 6542813024, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "BH-71Y-2851", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "38b55d8642902ac72f5987064e67fbe9b766a04dc338df11f6fa45d4cb64179e"}95{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "816ed2125ce7b04e762b5743", "attempt_id": "18ce15c66489", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:53.456831Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 437, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 240.14141701627523, "ttft_ms": 41.92666697781533, "decode_ms": 59.50945906806737}, "total_wall_ms": 342.6190000027418, "ttft_ms": 41.92666697781533, "input_usage": {"receiver_prompt_tokens": 437, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6279806976, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6537684812, "mlx_cache_bytes": 12875045005, "process_rss_bytes": 6279806976}, "peak_prefill_memory": null, "peak_decode_memory": 6537684812, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}96{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c71959c0d3e5357f7027f876", "attempt_id": "202e36495fb2", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:53.819679Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 435, "message_tokens": 344, "message_utf8_bytes": 817, "cache_tensor_bytes": null, "serialized_payload_bytes": 817, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29992.015082971193, "decode_payload_ms": 0.0, "receiver_prefill_ms": 189.47916699107736, "ttft_ms": 131.1209590639919, "decode_ms": 48.01504104398191}, "total_wall_ms": 371.0270000156015, "ttft_ms": 131.1209590639919, "input_usage": {"receiver_prompt_tokens": 435, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 940.3810419607908, "sender_ttft_ms": 44.13541697431356, "sender_decode_ms": 29004.778207978234}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6279839744, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6536767308, "mlx_cache_bytes": 13113464973, "process_rss_bytes": 6279839744}, "peak_prefill_memory": null, "peak_decode_memory": 6536767308, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 1, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "448fbefbbf3dc2a8c9785c435f1c717947c9d692820d830c64b908b542a8566e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}97{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ba2a8cdfd43ae8d3dda33750", "attempt_id": "eced51fba189", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:54.221815Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 141, "message_tokens": 51, "message_utf8_bytes": 117, "cache_tensor_bytes": null, "serialized_payload_bytes": 117, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4871.90762499813, "decode_payload_ms": 0.0, "receiver_prefill_ms": 163.6903330218047, "ttft_ms": 34.58358300849795, "decode_ms": 513.1103330058977}, "total_wall_ms": 5587.869332986884, "ttft_ms": 34.58358300849795, "input_usage": {"receiver_prompt_tokens": 141, "sender_input_tokens": 2123, "sender_output_tokens": 68, "sender_prefill_ms": 1078.6179170245305, "sender_ttft_ms": 71.23454206157476, "sender_decode_ms": 3716.0445830086246}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6239371264, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6295151728, "mlx_cache_bytes": 13235624077, "process_rss_bytes": 6239371264}, "peak_prefill_memory": null, "peak_decode_memory": 6295151728, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f3dabf50da33366b917608aeefa49c1007a0cc60466cdae60beb550e66756bbb", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}98{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "22d7c540fa1569284e76b10d", "attempt_id": "ed9984b05a3b", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:51:59.853948Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 145, "message_tokens": 49, "message_utf8_bytes": 117, "cache_tensor_bytes": null, "serialized_payload_bytes": 117, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4692.283415934071, "decode_payload_ms": 0.0, "receiver_prefill_ms": 220.4554999480024, "ttft_ms": 58.05037508253008, "decode_ms": 158.4277079673484}, "total_wall_ms": 5618.514208937995, "ttft_ms": 58.05037508253008, "input_usage": {"receiver_prompt_tokens": 145, "sender_input_tokens": 2129, "sender_output_tokens": 64, "sender_prefill_ms": 1026.6934579703957, "sender_ttft_ms": 56.53862503822893, "sender_decode_ms": 3594.4499170873314}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6215139328, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6298887280, "mlx_cache_bytes": 13292771469, "process_rss_bytes": 6215139328}, "peak_prefill_memory": null, "peak_decode_memory": 6298887280, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C in Schefferville", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "9652ad7c55f3009c5be7436cce78588b23912163734f44f8d5832818ead6c55f", "output_hash": "3491362e6ed6bb557ea2c25e29286d6fdbd0a0a36995d38446eddfd8450e097c"}99{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "74f83f0cb7ec6ed89fabe363", "attempt_id": "8d67161c6748", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:52:05.699446Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 146, "message_tokens": 50, "message_utf8_bytes": 115, "cache_tensor_bytes": null, "serialized_payload_bytes": 115, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3990.0535839842632, "decode_payload_ms": 0.0, "receiver_prefill_ms": 769.0995839657262, "ttft_ms": 57.462334050796926, "decode_ms": 316.9130419846624}, "total_wall_ms": 5153.69558299426, "ttft_ms": 57.462334050796926, "input_usage": {"receiver_prompt_tokens": 146, "sender_input_tokens": 2129, "sender_output_tokens": 61, "sender_prefill_ms": 1754.6524999197572, "sender_ttft_ms": 47.18887491617352, "sender_decode_ms": 2178.2806670526043}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 1690402816, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6299755632, "mlx_cache_bytes": 13292771469, "process_rss_bytes": 1690402816}, "peak_prefill_memory": null, "peak_decode_memory": 6299755632, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0484c7585de7098cb92373463b2c069927504a9946d31a9ca025d1a1167fabc5", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}100{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5a142f30cf64f0c904f3e59b", "attempt_id": "8adbee89466e", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:52:11.707981Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 141, "message_tokens": 48, "message_utf8_bytes": 118, "cache_tensor_bytes": null, "serialized_payload_bytes": 118, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5765.988125000149, "decode_payload_ms": 0.0, "receiver_prefill_ms": 608.8366659823805, "ttft_ms": 161.4062919979915, "decode_ms": 745.7209579879418}, "total_wall_ms": 7289.483958971687, "ttft_ms": 161.4062919979915, "input_usage": {"receiver_prompt_tokens": 141, "sender_input_tokens": 2126, "sender_output_tokens": 77, "sender_prefill_ms": 1012.2772080358118, "sender_ttft_ms": 57.683915947563946, "sender_decode_ms": 4690.014833933674}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 1691844608, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6295151728, "mlx_cache_bytes": 13320558733, "process_rss_bytes": 1691844608}, "peak_prefill_memory": null, "peak_decode_memory": 6295151728, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "904688de4b65b2f491fa2b715cc1c916e51e218dddc902c4585a7ed8ac38e786", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}101{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "39a2830786ba716d51b9e097", "attempt_id": "cb73da0dc216", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:52:19.059089Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 153, "message_tokens": 52, "message_utf8_bytes": 123, "cache_tensor_bytes": null, "serialized_payload_bytes": 123, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 8046.766332932748, "decode_payload_ms": 0.0, "receiver_prefill_ms": 412.3260419582948, "ttft_ms": 58.95604193210602, "decode_ms": 739.9513330310583}, "total_wall_ms": 9260.830290964805, "ttft_ms": 58.95604193210602, "input_usage": {"receiver_prompt_tokens": 153, "sender_input_tokens": 2134, "sender_output_tokens": 100, "sender_prefill_ms": 1188.6862500105053, "sender_ttft_ms": 54.72245905548334, "sender_decode_ms": 6797.676082933322}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 1693466624, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6305293424, "mlx_cache_bytes": 13420566669, "process_rss_bytes": 1693466624}, "peak_prefill_memory": null, "peak_decode_memory": 6305293424, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172 NOK", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.6666666666666666}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "8204a82f1decc01f3bb75bcfd65fcc9f2ca508a0bd791a8f43e1d8e066d0efb9", "output_hash": "ce5dae2b035af8d38bda9e7d1f1cb7eb7e4141e20298d86d8bce5fe5ac6c0246"}102{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5588aea7edf4c02460260f4a", "attempt_id": "b14f494e8798", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:52:28.357852Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 146, "message_tokens": 47, "message_utf8_bytes": 116, "cache_tensor_bytes": null, "serialized_payload_bytes": 116, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5077.3695829557255, "decode_payload_ms": 0.0, "receiver_prefill_ms": 142.33833295293152, "ttft_ms": 59.133582981303334, "decode_ms": 689.989916048944}, "total_wall_ms": 5974.0480419714, "ttft_ms": 59.133582981303334, "input_usage": {"receiver_prompt_tokens": 146, "sender_input_tokens": 2132, "sender_output_tokens": 64, "sender_prefill_ms": 1066.8570840498433, "sender_ttft_ms": 87.63541700318456, "sender_decode_ms": 3917.154208989814}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 1691222016, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6299755632, "mlx_cache_bytes": 13467342989, "process_rss_bytes": 1691222016}, "peak_prefill_memory": null, "peak_decode_memory": 6299755632, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "SP-38G-4217", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "1c4fa9d9ef27d880d8e001d842031994d04310a6e547ac6de726ad73dc66d03d", "output_hash": "e643c21a78559b96fd65325d9ce779b7b8fd6a47c53cf7c2cda1efa8412458c4"}103{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "bcd3acbe0f3d95a105aa30b2", "attempt_id": "eb28c3804423", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:52:34.375593Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 137, "message_tokens": 44, "message_utf8_bytes": 110, "cache_tensor_bytes": null, "serialized_payload_bytes": 110, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5784.818125073798, "decode_payload_ms": 0.0, "receiver_prefill_ms": 415.22275004535913, "ttft_ms": 34.29595800116658, "decode_ms": 36.267624935135245}, "total_wall_ms": 6273.845667019486, "ttft_ms": 34.29595800116658, "input_usage": {"receiver_prompt_tokens": 137, "sender_input_tokens": 2126, "sender_output_tokens": 66, "sender_prefill_ms": 1249.5219999691471, "sender_ttft_ms": 79.78362508583814, "sender_decode_ms": 4450.788166956045}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1690255360, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6291530864, "mlx_cache_bytes": 13467342989, "process_rss_bytes": 1690255360}, "peak_prefill_memory": null, "peak_decode_memory": 6291530864, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "8f10f95ff2ae53b376c8c654d82ff6f515acf8c6218364c4ab0e748c7680e868", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}104{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c3ba77cc9873d1268b158210", "attempt_id": "43c7d5ca78d9", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:52:40.687184Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 132, "message_tokens": 41, "message_utf8_bytes": 103, "cache_tensor_bytes": null, "serialized_payload_bytes": 103, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4641.5456250542775, "decode_payload_ms": 0.0, "receiver_prefill_ms": 265.1204589055851, "ttft_ms": 60.9997499268502, "decode_ms": 67.1607090625912}, "total_wall_ms": 5039.716250030324, "ttft_ms": 60.9997499268502, "input_usage": {"receiver_prompt_tokens": 132, "sender_input_tokens": 2124, "sender_output_tokens": 50, "sender_prefill_ms": 1234.726291964762, "sender_ttft_ms": 28.936250018887222, "sender_decode_ms": 3374.0929579362273}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1691385856, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6294971504, "mlx_cache_bytes": 13561518221, "process_rss_bytes": 1691385856}, "peak_prefill_memory": null, "peak_decode_memory": 6294971504, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "5fe520b019760ea0d31869b9808981ded50acb1dedef01a23bc802a10e688067", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}105{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "b469fc12ea505953e18a56f7", "attempt_id": "83e415ae1f39", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:21.405549Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1045, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 378.01750004291534, "ttft_ms": 47.55725001450628, "decode_ms": 145.22933401167393}, "total_wall_ms": 592.0917090261355, "ttft_ms": 47.55725001450628, "input_usage": {"receiver_prompt_tokens": 1045, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1437990912, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6818201660, "mlx_cache_bytes": 14222366861, "process_rss_bytes": 1437990912}, "peak_prefill_memory": null, "peak_decode_memory": 6818201660, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "561d3553701d834273477e7e81d6f4a2903e1fe9e09b5b4efdd73a721bde815e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}106{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "79d87ef074645777b892e823", "attempt_id": "6a8dab0a92ab", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:22.029246Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1051, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 541.6967920027673, "ttft_ms": 85.35870804917067, "decode_ms": 63.60841600690037}, "total_wall_ms": 693.0375420488417, "ttft_ms": 85.35870804917067, "input_usage": {"receiver_prompt_tokens": 1051, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1442496512, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6821904444, "mlx_cache_bytes": 14839896205, "process_rss_bytes": 1442496512}, "peak_prefill_memory": null, "peak_decode_memory": 6821904444, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ab0e9c7021e297c7521daf9f4fa54de8136f15dc9e1c13a37a4af71f57687c8e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}107{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0ae7cfeab51e1be141e559e2", "attempt_id": "7f47242b6bba", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:22.747613Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1051, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 494.5004169130698, "ttft_ms": 120.40316697675735, "decode_ms": 77.48079195152968}, "total_wall_ms": 694.4931669859216, "ttft_ms": 120.40316697675735, "input_usage": {"receiver_prompt_tokens": 1051, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1442250752, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6821904444, "mlx_cache_bytes": 14839896205, "process_rss_bytes": 1442250752}, "peak_prefill_memory": null, "peak_decode_memory": 6821904444, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ab0e9c7021e297c7521daf9f4fa54de8136f15dc9e1c13a37a4af71f57687c8e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}108{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "4c71988a2be93c8ed7c69a57", "attempt_id": "3a46e9dbd7eb", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:23.468816Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1048, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 461.9608750799671, "ttft_ms": 76.96991600096226, "decode_ms": 63.66766698192805}, "total_wall_ms": 605.1674999762326, "ttft_ms": 76.96991600096226, "input_usage": {"receiver_prompt_tokens": 1048, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1441071104, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6850789436, "mlx_cache_bytes": 15313885325, "process_rss_bytes": 1441071104}, "peak_prefill_memory": null, "peak_decode_memory": 6850789436, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ab0e9c7021e297c7521daf9f4fa54de8136f15dc9e1c13a37a4af71f57687c8e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}109{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9dca3b4c66d92aa3b42c1b1d", "attempt_id": "f69bd7f9a628", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:24.102955Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1056, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 458.21570802945644, "ttft_ms": 89.71616707276553, "decode_ms": 565.710791037418}, "total_wall_ms": 1115.8437500707805, "ttft_ms": 89.71616707276553, "input_usage": {"receiver_prompt_tokens": 1056, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 1440825344, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6824689724, "mlx_cache_bytes": 15934281869, "process_rss_bytes": 1440825344}, "peak_prefill_memory": null, "peak_decode_memory": 6824689724, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ab0e9c7021e297c7521daf9f4fa54de8136f15dc9e1c13a37a4af71f57687c8e", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}110{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f01b5e31b9d46d292e06da03", "attempt_id": "ec1a8cee9b17", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:25.242074Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1054, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 561.0457910224795, "ttft_ms": 73.62216606270522, "decode_ms": 76.63941592909396}, "total_wall_ms": 714.6227919729427, "ttft_ms": 73.62216606270522, "input_usage": {"receiver_prompt_tokens": 1054, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1442168832, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6823804988, "mlx_cache_bytes": 16367802509, "process_rss_bytes": 1442168832}, "peak_prefill_memory": null, "peak_decode_memory": 6823804988, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "561d3553701d834273477e7e81d6f4a2903e1fe9e09b5b4efdd73a721bde815e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}111{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "360276a13afb2d41417e890d", "attempt_id": "96c0dbfa9158", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:25.983297Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1048, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 497.2743750549853, "ttft_ms": 80.10491693858057, "decode_ms": 52.842916920781136}, "total_wall_ms": 632.7891250839457, "ttft_ms": 80.10491693858057, "input_usage": {"receiver_prompt_tokens": 1048, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1442299904, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6855032892, "mlx_cache_bytes": 16367802509, "process_rss_bytes": 1442299904}, "peak_prefill_memory": null, "peak_decode_memory": 6855032892, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ab0e9c7021e297c7521daf9f4fa54de8136f15dc9e1c13a37a4af71f57687c8e", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}112{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "e908282eb7cc900972572715", "attempt_id": "eab2dcbcb36e", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:26.640776Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 7}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 1046, "message_tokens": 970, "message_utf8_bytes": 623, "cache_tensor_bytes": null, "serialized_payload_bytes": 623, "side_information_bytes": 0, "fetched_bytes": 1634, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 35642.67829200253, "decode_payload_ms": 0.0, "receiver_prefill_ms": 502.8000839520246, "ttft_ms": 92.95654203742743, "decode_ms": 70.77137497253716}, "total_wall_ms": 669.1975829889998, "ttft_ms": 92.95654203742743, "input_usage": {"receiver_prompt_tokens": 1046, "sender_input_tokens": 2115, "sender_output_tokens": 512, "sender_prefill_ms": 1526.9413749920204, "sender_ttft_ms": 87.4640840338543, "sender_decode_ms": 34020.01454192214}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 1442349056, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6854344764, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 1442349056}, "peak_prefill_memory": null, "peak_decode_memory": 6854344764, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "no", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1634, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ab0e9c7021e297c7521daf9f4fa54de8136f15dc9e1c13a37a4af71f57687c8e", "output_hash": "9390298f3fb0c5b160498935d79cb139aef28e1c47358b4bbba61862b9c26e59"}113{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ac30ac2b8b432c7b33420750", "attempt_id": "8d3b43d971cd", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:27.376585Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2045, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 4354.247833020054, "ttft_ms": 231.00641707424074, "decode_ms": 1004.8636249266565}, "total_wall_ms": 5597.181500052102, "ttft_ms": 231.00641707424074, "input_usage": {"receiver_prompt_tokens": 2045}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 245317632, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7061651504, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 245317632}, "peak_prefill_memory": null, "peak_decode_memory": 7061651504, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}114{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2713cf06493be2d1b061fee3", "attempt_id": "c0d0f989295e", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:32.996839Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2051, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 4630.934291984886, "ttft_ms": 23.955959011800587, "decode_ms": 422.55058290902525}, "total_wall_ms": 5082.146625034511, "ttft_ms": 23.955959011800587, "input_usage": {"receiver_prompt_tokens": 2051}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 56459264, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054786612, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 56459264}, "peak_prefill_memory": null, "peak_decode_memory": 7054786612, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}115{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "919ff6dc3f9281577d51d172", "attempt_id": "ee701fe8b615", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:38.106049Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2051, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 848.609208012931, "ttft_ms": 24.180709035135806, "decode_ms": 170.95233302097768}, "total_wall_ms": 1054.8324170522392, "ttft_ms": 24.180709035135806, "input_usage": {"receiver_prompt_tokens": 2051}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 57016320, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054786612, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 57016320}, "peak_prefill_memory": null, "peak_decode_memory": 7054786612, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}116{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0f36651cea9cff68239c7838", "attempt_id": "47d568264d49", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:39.193348Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2048, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 845.6971660489216, "ttft_ms": 41.79829196073115, "decode_ms": 663.4332499234006}, "total_wall_ms": 1657.8844169853255, "ttft_ms": 41.79829196073115, "input_usage": {"receiver_prompt_tokens": 2048}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 47415296, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7063175232, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 47415296}, "peak_prefill_memory": null, "peak_decode_memory": 7063175232, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}117{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "1e9b39b54393c0c1c02f47b8", "attempt_id": "c165df7325ab", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:40.890183Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2056, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 690.7639999408275, "ttft_ms": 41.85066593345255, "decode_ms": 357.3740000138059}, "total_wall_ms": 1096.705208066851, "ttft_ms": 41.85066593345255, "input_usage": {"receiver_prompt_tokens": 2056}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 57409536, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054792797, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 57409536}, "peak_prefill_memory": null, "peak_decode_memory": 7054792797, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}118{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "cd633990feda5f0bef12034c", "attempt_id": "635746671566", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:42.112634Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q05_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2054, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 779.1565000079572, "ttft_ms": 53.26924996916205, "decode_ms": 484.53966702800244}, "total_wall_ms": 1322.8936250088736, "ttft_ms": 53.26924996916205, "input_usage": {"receiver_prompt_tokens": 2054}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 56623104, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7054792797, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 56623104}, "peak_prefill_memory": null, "peak_decode_memory": 7054792797, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "SP-38G-4217", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "e643c21a78559b96fd65325d9ce779b7b8fd6a47c53cf7c2cda1efa8412458c4"}119{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "91aee09f95fdd9157afefac1", "attempt_id": "cb67f30319df", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:43.460009Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q06_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2048, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 760.179084027186, "ttft_ms": 47.706791083328426, "decode_ms": 55.15999998897314}, "total_wall_ms": 866.7989170644432, "ttft_ms": 47.706791083328426, "input_usage": {"receiver_prompt_tokens": 2048}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 57098240, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7063175232, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 57098240}, "peak_prefill_memory": null, "peak_decode_memory": 7063175232, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}120{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "174b32ecb6104665ab692120", "attempt_id": "f451067dc38f", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "d7fbbfa78a1b00df1963068522ef3d136b604128276634399fd3eee67f781637", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:53:44.350599Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q07_negation", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 8, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 2046, "message_tokens": 1971, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 39.819792029447854, "decode_payload_ms": 0.29975001234561205, "receiver_prefill_ms": 2192.5274999812245, "ttft_ms": 74.35687503311783, "decode_ms": 55.369207984767854}, "total_wall_ms": 2326.2041250709444, "ttft_ms": 74.35687503311783, "input_usage": {"receiver_prompt_tokens": 2046}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 57180160, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7062388800, "mlx_cache_bytes": 16798177421, "process_rss_bytes": 57180160}, "peak_prefill_memory": null, "peak_decode_memory": 7062388800, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "No", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "1ea442a134b2a184bd5d40104401f2a37fbc09ccf3f4bc9da161c6099be3691d"}121{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "279d8b722165cabc9226b358", "attempt_id": "d58f4302ef5e", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:55:01.338876Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.2597666159309995, "origin_tokens": 1766, "compressed_tokens": 552, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 749, "message_tokens": 673, "message_utf8_bytes": 1901, "cache_tensor_bytes": null, "serialized_payload_bytes": 1901, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 32285.182291991077, "decode_payload_ms": 0.0, "receiver_prefill_ms": 24393.165249959566, "ttft_ms": 20765.287874964997, "decode_ms": 80176.83241702616}, "total_wall_ms": 125422.81791707501, "ttft_ms": 20765.287874964997, "input_usage": {"receiver_prompt_tokens": 749}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 204587008, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6651358216, "mlx_cache_bytes": 492087292, "process_rss_bytes": 204587008}, "peak_prefill_memory": null, "peak_decode_memory": 6651358216, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "de22f94c4433681b0a927b8c9becabd6a356dc8451924add5adb90472d4a9bdc", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}122{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "966e0a4e843f1f6f8e0409ab", "attempt_id": "3b2fa3c786fb", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:57:06.875896Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.2597666159309995, "origin_tokens": 1766, "compressed_tokens": 552, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 755, "message_tokens": 673, "message_utf8_bytes": 1901, "cache_tensor_bytes": null, "serialized_payload_bytes": 1901, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 32285.182291991077, "decode_payload_ms": 0.0, "receiver_prefill_ms": 15566.239541047253, "ttft_ms": 21677.49995808117, "decode_ms": 85318.63916595466}, "total_wall_ms": 122608.2591660088, "ttft_ms": 21677.49995808117, "input_usage": {"receiver_prompt_tokens": 755}, "output_usage": {"output_tokens": 8}, "provider_cached_usage": null, "process_memory": 47808512, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6655291480, "mlx_cache_bytes": 951363580, "process_rss_bytes": 47808512}, "peak_prefill_memory": null, "peak_decode_memory": 6655291480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "de22f94c4433681b0a927b8c9becabd6a356dc8451924add5adb90472d4a9bdc", "output_hash": "620632e3a4d38128fdc81039176e9ca7b44bb1593e3f778652709568225a772b"}123{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2e4429c4fac16489b3bee409", "attempt_id": "e605803a1203", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:09.553196Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.2597666159309995, "origin_tokens": 1766, "compressed_tokens": 552, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 755, "message_tokens": 673, "message_utf8_bytes": 1901, "cache_tensor_bytes": null, "serialized_payload_bytes": 1901, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 32285.182291991077, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1988.4534999728203, "ttft_ms": 56.966249947436154, "decode_ms": 461.8136660428718}, "total_wall_ms": 2518.857416929677, "ttft_ms": 56.966249947436154, "input_usage": {"receiver_prompt_tokens": 755}, "output_usage": {"output_tokens": 8}, "provider_cached_usage": null, "process_memory": 63848448, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6655291480, "mlx_cache_bytes": 951363580, "process_rss_bytes": 63848448}, "peak_prefill_memory": null, "peak_decode_memory": 6655291480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "71. 2449", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "de22f94c4433681b0a927b8c9becabd6a356dc8451924add5adb90472d4a9bdc", "output_hash": "10a9b36bf447655a418edd0d42d693f04d09dc9b21e23caa5ac3606fade43800"}124{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "70c2ceb5ce8ff3b887c236ca", "attempt_id": "574a02ba0e67", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:12.137951Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.2597666159309995, "origin_tokens": 1766, "compressed_tokens": 552, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 752, "message_tokens": 673, "message_utf8_bytes": 1901, "cache_tensor_bytes": null, "serialized_payload_bytes": 1901, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 32285.182291991077, "decode_payload_ms": 0.0, "receiver_prefill_ms": 364.38358295708895, "ttft_ms": 52.571208914741874, "decode_ms": 2679.71958394628}, "total_wall_ms": 3101.1328749591485, "ttft_ms": 52.571208914741874, "input_usage": {"receiver_prompt_tokens": 752}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 65290240, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6669299788, "mlx_cache_bytes": 1334716412, "process_rss_bytes": 65290240}, "peak_prefill_memory": null, "peak_decode_memory": 6669299788, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Eshtal-12 proposed Oksana Okonkwo October 2030 budget, 336. February 2031 revised, 767.2031 budget Eshtal", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "de22f94c4433681b0a927b8c9becabd6a356dc8451924add5adb90472d4a9bdc", "output_hash": "f2005847552db685445106002595bfeea7d3cc8652091b8366900a4888c8b62d"}125{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2b2b1c40fcff26d17ef62b15", "attempt_id": "701ad57adecc", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:15.266736Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.2597666159309995, "origin_tokens": 1766, "compressed_tokens": 552, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 760, "message_tokens": 673, "message_utf8_bytes": 1901, "cache_tensor_bytes": null, "serialized_payload_bytes": 1901, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 32285.182291991077, "decode_payload_ms": 0.0, "receiver_prefill_ms": 426.6626660246402, "ttft_ms": 86.61887503694743, "decode_ms": 169.83833292033523}, "total_wall_ms": 684.7365000285208, "ttft_ms": 86.61887503694743, "input_usage": {"receiver_prompt_tokens": 760}, "output_usage": {"output_tokens": 3}, "provider_cached_usage": null, "process_memory": 65486848, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6657896536, "mlx_cache_bytes": 1777969148, "process_rss_bytes": 65486848}, "peak_prefill_memory": null, "peak_decode_memory": 6657896536, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "889", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "de22f94c4433681b0a927b8c9becabd6a356dc8451924add5adb90472d4a9bdc", "output_hash": "a829c72c42755e384141ad8f163e4965ef5c9f8f0e07378c1d05a7222af0dd80"}126{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "acaabc6e4d4c49b34962e380", "attempt_id": "634c79534f08", "run_group_id": "E1_pilot:B0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:17.030978Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112716, "side_information_bytes": 16076, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1054.6357079874724, "doc_prefill_ms": 978.4983330173418, "compress_ms": 12.819958967156708, "serialize_ms": 51.819249987602234, "load_ms": 4.219291964545846, "receiver_prefill_ms": 0.0, "query_prefill_ms": 88.34758296143264, "ttft_ms": 88.34758296143264, "decode_ms": 593.2706659659743}, "total_wall_ms": 688.8123749522492, "ttft_ms": 88.34758296143264, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 220413952, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6443311308, "mlx_cache_bytes": 2791001414, "process_rss_bytes": 220413952}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6443311308, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0fd1e41ac9de2a394237f6d73e79f07db7814fc3daad26bdafb5187f0b9128ae", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}127{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8a936ce40733cee79b7ce20f", "attempt_id": "44ff31a847c1", "run_group_id": "E1_pilot:B0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:17.743323Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112716, "side_information_bytes": 16076, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1054.6357079874724, "doc_prefill_ms": 978.4983330173418, "compress_ms": 12.819958967156708, "serialize_ms": 51.819249987602234, "load_ms": 3.2108330633491278, "receiver_prefill_ms": 0.0, "query_prefill_ms": 82.17745902948081, "ttft_ms": 82.17745902948081, "decode_ms": 468.55266601778567}, "total_wall_ms": 557.0991659769788, "ttft_ms": 82.17745902948081, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 220889088, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6450127076, "mlx_cache_bytes": 2831355534, "process_rss_bytes": 220889088}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6450127076, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0fd1e41ac9de2a394237f6d73e79f07db7814fc3daad26bdafb5187f0b9128ae", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}128{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "77373cac251170265921d9e9", "attempt_id": "24817c9f86d9", "run_group_id": "E1_pilot:B0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:18.327873Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112716, "side_information_bytes": 16076, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1054.6357079874724, "doc_prefill_ms": 978.4983330173418, "compress_ms": 12.819958967156708, "serialize_ms": 51.819249987602234, "load_ms": 1.7341249622404575, "receiver_prefill_ms": 0.0, "query_prefill_ms": 62.84120795316994, "ttft_ms": 62.84120795316994, "decode_ms": 408.66533399093896}, "total_wall_ms": 482.51424997579306, "ttft_ms": 62.84120795316994, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 221806592, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6450127076, "mlx_cache_bytes": 2831355534, "process_rss_bytes": 221806592}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6450127076, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0fd1e41ac9de2a394237f6d73e79f07db7814fc3daad26bdafb5187f0b9128ae", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}129{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "e9dbabb20e5ecf6825c093f9", "attempt_id": "73ed99a053fb", "run_group_id": "E1_pilot:B0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:18.831982Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112716, "side_information_bytes": 16076, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1054.6357079874724, "doc_prefill_ms": 978.4983330173418, "compress_ms": 12.819958967156708, "serialize_ms": 51.819249987602234, "load_ms": 1.9901669584214687, "receiver_prefill_ms": 0.0, "query_prefill_ms": 418.13612496480346, "ttft_ms": 418.13612496480346, "decode_ms": 330.44787496328354}, "total_wall_ms": 752.8034159913659, "ttft_ms": 418.13612496480346, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 222478336, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6448701668, "mlx_cache_bytes": 2860700722, "process_rss_bytes": 222478336}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6448701668, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0fd1e41ac9de2a394237f6d73e79f07db7814fc3daad26bdafb5187f0b9128ae", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}130{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "1d1e50b0f0e309dfdab391d3", "attempt_id": "31f55af6e6be", "run_group_id": "E1_pilot:B0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:19.611557Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112716, "side_information_bytes": 16076, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1054.6357079874724, "doc_prefill_ms": 978.4983330173418, "compress_ms": 12.819958967156708, "serialize_ms": 51.819249987602234, "load_ms": 1.3780000153928995, "receiver_prefill_ms": 0.0, "query_prefill_ms": 67.90320901200175, "ttft_ms": 67.90320901200175, "decode_ms": 415.6412499723956}, "total_wall_ms": 486.8782910052687, "ttft_ms": 67.90320901200175, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 222527488, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6454108408, "mlx_cache_bytes": 2903135650, "process_rss_bytes": 222527488}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6454108408, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0fd1e41ac9de2a394237f6d73e79f07db7814fc3daad26bdafb5187f0b9128ae", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}131{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "b6c460332a7cf3d5fadc37fa", "attempt_id": "4012c407ee61", "run_group_id": "E1_pilot:B1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:20.150779Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": "5934e1e7b2bd92f6", "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112744, "side_information_bytes": 16104, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29.612625017762184, "doc_prefill_ms": 978.4983330173418, "compress_ms": 2.880375017412007, "serialize_ms": 26.52341697830707, "load_ms": 37.262209109030664, "receiver_prefill_ms": 0.0, "query_prefill_ms": 86.25749999191612, "ttft_ms": 86.25749999191612, "decode_ms": 548.937665997073}, "total_wall_ms": 674.4144590338692, "ttft_ms": 86.25749999191612, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 297304064, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6443311308, "mlx_cache_bytes": 2903135650, "process_rss_bytes": 297304064}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6443311308, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "eb4d9d05bd93beebadeb0b446aca6f8cd3f69a85009c57c2686509661e53bad1", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}132{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "14d5e1a8f0c5f483f2ca4ce7", "attempt_id": "1c6f599033bc", "run_group_id": "E1_pilot:B1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:20.850633Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": "5934e1e7b2bd92f6", "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112744, "side_information_bytes": 16104, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29.612625017762184, "doc_prefill_ms": 978.4983330173418, "compress_ms": 2.880375017412007, "serialize_ms": 26.52341697830707, "load_ms": 15.384000027552247, "receiver_prefill_ms": 0.0, "query_prefill_ms": 88.41804193798453, "ttft_ms": 88.41804193798453, "decode_ms": 505.07045804988593}, "total_wall_ms": 611.153458012268, "ttft_ms": 88.41804193798453, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 297353216, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6450127076, "mlx_cache_bytes": 2903135650, "process_rss_bytes": 297353216}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6450127076, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "eb4d9d05bd93beebadeb0b446aca6f8cd3f69a85009c57c2686509661e53bad1", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}133{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ea781cec6fc361596c57c44b", "attempt_id": "b6d981c46bf5", "run_group_id": "E1_pilot:B1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:21.490666Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": "5934e1e7b2bd92f6", "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112744, "side_information_bytes": 16104, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29.612625017762184, "doc_prefill_ms": 978.4983330173418, "compress_ms": 2.880375017412007, "serialize_ms": 26.52341697830707, "load_ms": 12.7028749557212, "receiver_prefill_ms": 0.0, "query_prefill_ms": 87.66562503296882, "ttft_ms": 87.66562503296882, "decode_ms": 407.7478339895606}, "total_wall_ms": 510.0972499931231, "ttft_ms": 87.66562503296882, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 297418752, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6450127076, "mlx_cache_bytes": 2903135650, "process_rss_bytes": 297418752}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6450127076, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "eb4d9d05bd93beebadeb0b446aca6f8cd3f69a85009c57c2686509661e53bad1", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}134{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "b5ec1737c47385416e969667", "attempt_id": "04d7bcbbb57d", "run_group_id": "E1_pilot:B1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:22.024532Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": "5934e1e7b2bd92f6", "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112744, "side_information_bytes": 16104, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29.612625017762184, "doc_prefill_ms": 978.4983330173418, "compress_ms": 2.880375017412007, "serialize_ms": 26.52341697830707, "load_ms": 14.886916964314878, "receiver_prefill_ms": 0.0, "query_prefill_ms": 73.55716696474701, "ttft_ms": 73.55716696474701, "decode_ms": 298.2067910488695}, "total_wall_ms": 388.62062501721084, "ttft_ms": 73.55716696474701, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 297451520, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6448586980, "mlx_cache_bytes": 2903135650, "process_rss_bytes": 297451520}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6448586980, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "eb4d9d05bd93beebadeb0b446aca6f8cd3f69a85009c57c2686509661e53bad1", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}135{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0979bdfd41adddd70a3c17ae", "attempt_id": "ad0d6612c826", "run_group_id": "E1_pilot:B1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:22.441259Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6321692680}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": "5934e1e7b2bd92f6", "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112744, "side_information_bytes": 16104, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 29.612625017762184, "doc_prefill_ms": 978.4983330173418, "compress_ms": 2.880375017412007, "serialize_ms": 26.52341697830707, "load_ms": 13.400583993643522, "receiver_prefill_ms": 0.0, "query_prefill_ms": 112.01254196930677, "ttft_ms": 112.01254196930677, "decode_ms": 404.5315410476178}, "total_wall_ms": 531.9922911003232, "ttft_ms": 112.01254196930677, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 297467904, "backend_memory": {"mlx_active_bytes": 6398369800, "mlx_peak_bytes": 6454108408, "mlx_cache_bytes": 2903135650, "process_rss_bytes": 297467904}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6454108408, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "eb4d9d05bd93beebadeb0b446aca6f8cd3f69a85009c57c2686509661e53bad1", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}136{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "d332386db6784efc64c1d901", "attempt_id": "89423c9b6067", "run_group_id": "E1_pilot:B2|bits=8", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:23.079020Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6287439112}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 39363840, "serialized_payload_bytes": 39398030, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 81.0227079782635, "doc_prefill_ms": 978.4983330173418, "compress_ms": 62.38841696176678, "serialize_ms": 17.83695793710649, "load_ms": 10.063333087600768, "receiver_prefill_ms": 0.0, "query_prefill_ms": 98.01379207056016, "ttft_ms": 98.01379207056016, "decode_ms": 642.7352499449626}, "total_wall_ms": 752.6289579691365, "ttft_ms": 98.01379207056016, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 344621056, "backend_memory": {"mlx_active_bytes": 6331086088, "mlx_peak_bytes": 6385244376, "mlx_cache_bytes": 3107838578, "process_rss_bytes": 344621056}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6385244376, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "5f8624f35d651cd19bdb054c3205083465b76bee66103dc2a4d23dbd2e6dc167", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}137{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ce8c36602a602b283ae79103", "attempt_id": "d5a56b03124d", "run_group_id": "E1_pilot:B2|bits=8", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:23.856723Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6287439112}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 39363840, "serialized_payload_bytes": 39398030, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 81.0227079782635, "doc_prefill_ms": 978.4983330173418, "compress_ms": 62.38841696176678, "serialize_ms": 17.83695793710649, "load_ms": 1.7602919833734632, "receiver_prefill_ms": 0.0, "query_prefill_ms": 73.62908299546689, "ttft_ms": 73.62908299546689, "decode_ms": 581.0134579660371}, "total_wall_ms": 658.16533297766, "ttft_ms": 73.62908299546689, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 344866816, "backend_memory": {"mlx_active_bytes": 6331086088, "mlx_peak_bytes": 6397341488, "mlx_cache_bytes": 3132174722, "process_rss_bytes": 344866816}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6397341488, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "5f8624f35d651cd19bdb054c3205083465b76bee66103dc2a4d23dbd2e6dc167", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}138{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6dadda20b29533e19a33cf18", "attempt_id": "e7f2c16a7620", "run_group_id": "E1_pilot:B2|bits=8", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:24.537804Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6287439112}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 39363840, "serialized_payload_bytes": 39398030, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 81.0227079782635, "doc_prefill_ms": 978.4983330173418, "compress_ms": 62.38841696176678, "serialize_ms": 17.83695793710649, "load_ms": 4.641667008399963, "receiver_prefill_ms": 0.0, "query_prefill_ms": 87.89008401799947, "ttft_ms": 87.89008401799947, "decode_ms": 413.85362506844103}, "total_wall_ms": 508.1339169992134, "ttft_ms": 87.89008401799947, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 344883200, "backend_memory": {"mlx_active_bytes": 6331086088, "mlx_peak_bytes": 6397342960, "mlx_cache_bytes": 3132174722, "process_rss_bytes": 344883200}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6397342960, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "5f8624f35d651cd19bdb054c3205083465b76bee66103dc2a4d23dbd2e6dc167", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}139{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "e700fa2d22c5d68a0eff00a3", "attempt_id": "ece0de4ceb84", "run_group_id": "E1_pilot:B2|bits=8", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:25.069353Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6287439112}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 39363840, "serialized_payload_bytes": 39398030, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 81.0227079782635, "doc_prefill_ms": 978.4983330173418, "compress_ms": 62.38841696176678, "serialize_ms": 17.83695793710649, "load_ms": 3.1166670378297567, "receiver_prefill_ms": 0.0, "query_prefill_ms": 223.1630829628557, "ttft_ms": 223.1630829628557, "decode_ms": 311.76733307074755}, "total_wall_ms": 539.6941249491647, "ttft_ms": 223.1630829628557, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 345047040, "backend_memory": {"mlx_active_bytes": 6331086088, "mlx_peak_bytes": 6391805168, "mlx_cache_bytes": 3156226434, "process_rss_bytes": 345047040}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6391805168, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "5f8624f35d651cd19bdb054c3205083465b76bee66103dc2a4d23dbd2e6dc167", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}140{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a5acf6de255115fd7dc04e29", "attempt_id": "385f04d3e231", "run_group_id": "E1_pilot:B2|bits=8", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:25.632423Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6287439112}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 39363840, "serialized_payload_bytes": 39398030, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 81.0227079782635, "doc_prefill_ms": 978.4983330173418, "compress_ms": 62.38841696176678, "serialize_ms": 17.83695793710649, "load_ms": 2.546416944824159, "receiver_prefill_ms": 0.0, "query_prefill_ms": 79.18208301998675, "ttft_ms": 79.18208301998675, "decode_ms": 429.49112493079156}, "total_wall_ms": 512.9653329495341, "ttft_ms": 79.18208301998675, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 345063424, "backend_memory": {"mlx_active_bytes": 6331086088, "mlx_peak_bytes": 6403224084, "mlx_cache_bytes": 3184888674, "process_rss_bytes": 345063424}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6403224084, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "5f8624f35d651cd19bdb054c3205083465b76bee66103dc2a4d23dbd2e6dc167", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}141{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "7cd936a2450994b0faf02135", "attempt_id": "491b7b0c189d", "run_group_id": "E1_pilot:B2|bits=4", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:26.238317Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6268564744}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 4, "realized": {"bits": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 20839680, "serialized_payload_bytes": 20873736, "side_information_bytes": 34056, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 67.32925004325807, "doc_prefill_ms": 978.4983330173418, "compress_ms": 55.0057920627296, "serialize_ms": 11.838207952678204, "load_ms": 4.002374946139753, "receiver_prefill_ms": 0.0, "query_prefill_ms": 67.90645804721862, "ttft_ms": 67.90645804721862, "decode_ms": 2813.5319590801373}, "total_wall_ms": 2887.6000830205157, "ttft_ms": 67.90645804721862, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 366903296, "backend_memory": {"mlx_active_bytes": 6293337352, "mlx_peak_bytes": 6343993640, "mlx_cache_bytes": 3290173922, "process_rss_bytes": 366903296}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6343993640, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "\n\nunknownunknownunknown memo\n\n5\nunknown memo\n5\n\n\n memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo memo", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "76d58caa4868c5c31b8703292423da4d64fc4061ef7af423b7b0c2b1f80e4f14", "output_hash": "78d55bf7eaf8f5d5718fae1592c5390a9d4187e68a847a5372b8f5fb9e43768c"}142{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "94492e56572670a9559d73c6", "attempt_id": "f59fbcc64611", "run_group_id": "E1_pilot:B2|bits=4", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:29.153268Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6268564744}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 4, "realized": {"bits": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 20839680, "serialized_payload_bytes": 20873736, "side_information_bytes": 34056, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 67.32925004325807, "doc_prefill_ms": 978.4983330173418, "compress_ms": 55.0057920627296, "serialize_ms": 11.838207952678204, "load_ms": 3.526707994751632, "receiver_prefill_ms": 0.0, "query_prefill_ms": 80.7841670466587, "ttft_ms": 80.7841670466587, "decode_ms": 2763.325000065379}, "total_wall_ms": 2849.4511670432985, "ttft_ms": 80.7841670466587, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 367132672, "backend_memory": {"mlx_active_bytes": 6293337352, "mlx_peak_bytes": 6355811184, "mlx_cache_bytes": 3290173922, "process_rss_bytes": 367132672}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6355811184, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": " \nuser\n\n memo20.\nat000000000 of000000000000000000000000 of000", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "76d58caa4868c5c31b8703292423da4d64fc4061ef7af423b7b0c2b1f80e4f14", "output_hash": "f956396894d647df89bc96127bf3812773aa370217fc6449ccdda4300087a89f"}143{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "853a5b240bffe439cd749368", "attempt_id": "04c8da66ebbb", "run_group_id": "E1_pilot:B2|bits=4", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:32.028236Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6268564744}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 4, "realized": {"bits": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 20839680, "serialized_payload_bytes": 20873736, "side_information_bytes": 34056, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 67.32925004325807, "doc_prefill_ms": 978.4983330173418, "compress_ms": 55.0057920627296, "serialize_ms": 11.838207952678204, "load_ms": 6.513500004075468, "receiver_prefill_ms": 0.0, "query_prefill_ms": 62.48629197943956, "ttft_ms": 62.48629197943956, "decode_ms": 106.45483306143433}, "total_wall_ms": 177.16833390295506, "ttft_ms": 62.48629197943956, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 367230976, "backend_memory": {"mlx_active_bytes": 6293337352, "mlx_peak_bytes": 6355811184, "mlx_cache_bytes": 3290173922, "process_rss_bytes": 367230976}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6355811184, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "\n\n", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "invalid_output", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "76d58caa4868c5c31b8703292423da4d64fc4061ef7af423b7b0c2b1f80e4f14", "output_hash": "75a11da44c802486bc6f65640aa48a730f0f684c5c07a42ba3cd1735eb3fb070"}144{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "81f140027c0a997402ce4f2e", "attempt_id": "7f67524fc3a1", "run_group_id": "E1_pilot:B2|bits=4", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:32.229035Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6268564744}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 4, "realized": {"bits": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 20839680, "serialized_payload_bytes": 20873736, "side_information_bytes": 34056, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 67.32925004325807, "doc_prefill_ms": 978.4983330173418, "compress_ms": 55.0057920627296, "serialize_ms": 11.838207952678204, "load_ms": 5.260500009171665, "receiver_prefill_ms": 0.0, "query_prefill_ms": 190.32929092645645, "ttft_ms": 190.32929092645645, "decode_ms": 2674.499708926305}, "total_wall_ms": 2871.8064579879865, "ttft_ms": 190.32929092645645, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 367476736, "backend_memory": {"mlx_active_bytes": 6293337352, "mlx_peak_bytes": 6351125360, "mlx_cache_bytes": 3290173922, "process_rss_bytes": 367476736}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6351125360, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "user\n\nsystem 0 0 000000000000000000000000000000000000000", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "76d58caa4868c5c31b8703292423da4d64fc4061ef7af423b7b0c2b1f80e4f14", "output_hash": "d3da647a46cbd3eb57b7ba22bfc90fff80e5d0dfb8efe06d0ea15bf1b7249bde"}145{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a72bd27a56291013f5fe2876", "attempt_id": "e5a7713892f4", "run_group_id": "E1_pilot:B2|bits=4", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:35.125847Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6268564744}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 4, "realized": {"bits": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 20839680, "serialized_payload_bytes": 20873736, "side_information_bytes": 34056, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 67.32925004325807, "doc_prefill_ms": 978.4983330173418, "compress_ms": 55.0057920627296, "serialize_ms": 11.838207952678204, "load_ms": 3.285999991931021, "receiver_prefill_ms": 0.0, "query_prefill_ms": 82.0221669273451, "ttft_ms": 82.0221669273451, "decode_ms": 2893.6651669209823}, "total_wall_ms": 2980.708874994889, "ttft_ms": 82.0221669273451, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 367493120, "backend_memory": {"mlx_active_bytes": 6293337352, "mlx_peak_bytes": 6361696052, "mlx_cache_bytes": 3290173922, "process_rss_bytes": 367493120}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6361696052, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "\n\n\n\n\n\n\n\n\n1 5 4 0 2 2 2 2 \n the the the the the the the the the the the the the the", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "76d58caa4868c5c31b8703292423da4d64fc4061ef7af423b7b0c2b1f80e4f14", "output_hash": "e8bbb8477ac2e890e7a22ce3d59d8e25dde723ba6f783fb1cd5c1e25d71bd090"}146{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5824d534bcfbd72198cf99b4", "attempt_id": "56c5a44010c0", "run_group_id": "E1_pilot:B2|bits=2|group_size=32", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:38.201954Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6262415368}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 2, "group_size": 32, "realized": {"bits": 2}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 13893120, "serialized_payload_bytes": 13926962, "side_information_bytes": 33842, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 70.37537498399615, "doc_prefill_ms": 978.4983330173418, "compress_ms": 56.51024996768683, "serialize_ms": 13.307290966622531, "load_ms": 3.6454590735957026, "receiver_prefill_ms": 0.0, "query_prefill_ms": 251.5014170203358, "ttft_ms": 251.5014170203358, "decode_ms": 2870.899665984325}, "total_wall_ms": 3129.833917017095, "ttft_ms": 251.5014170203358, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 379731968, "backend_memory": {"mlx_active_bytes": 6280634376, "mlx_peak_bytes": 6329204264, "mlx_cache_bytes": 3336158306, "process_rss_bytes": 379731968}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6329204264, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "eso\nñailed in the project the later the\nThe-he is 1hehehe the the hee short Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health Health", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7960b1fa79fbc02499ce6229a9a9eec92f86bde6fcca0fb9a9948b61cba15e6f", "output_hash": "ffe9e1c526a82a08613608e4c4f218dff9d7359341ef3b519fc07c5035dcc421"}147{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "41f2163c35c44d7c549109a9", "attempt_id": "0c3331abb4c3", "run_group_id": "E1_pilot:B2|bits=2|group_size=32", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:41.358257Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6262415368}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 2, "group_size": 32, "realized": {"bits": 2}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 13893120, "serialized_payload_bytes": 13926962, "side_information_bytes": 33842, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 70.37537498399615, "doc_prefill_ms": 978.4983330173418, "compress_ms": 56.51024996768683, "serialize_ms": 13.307290966622531, "load_ms": 4.8314169980585575, "receiver_prefill_ms": 0.0, "query_prefill_ms": 80.85762499831617, "ttft_ms": 80.85762499831617, "decode_ms": 2783.167792018503}, "total_wall_ms": 2870.5562920076773, "ttft_ms": 80.85762499831617, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 379961344, "backend_memory": {"mlx_active_bytes": 6280634376, "mlx_peak_bytes": 6341282553, "mlx_cache_bytes": 3336182947, "process_rss_bytes": 379961344}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6341282553, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": " \n\n0\n2\n2\n2\n2\n2\n2\n\n0\n4\n of the of the of the of the of the of the of the of the of the of the of the of the the of the", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7960b1fa79fbc02499ce6229a9a9eec92f86bde6fcca0fb9a9948b61cba15e6f", "output_hash": "c1ca748f56503d4863f9966df300c79baf9d9456236b1cf2950e04720a719536"}148{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0854d0e9d9ef7715ad3b0bb0", "attempt_id": "b06d89430872", "run_group_id": "E1_pilot:B2|bits=2|group_size=32", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:44.253780Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6262415368}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 2, "group_size": 32, "realized": {"bits": 2}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 13893120, "serialized_payload_bytes": 13926962, "side_information_bytes": 33842, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 70.37537498399615, "doc_prefill_ms": 978.4983330173418, "compress_ms": 56.51024996768683, "serialize_ms": 13.307290966622531, "load_ms": 3.1506670638918877, "receiver_prefill_ms": 0.0, "query_prefill_ms": 56.85079097747803, "ttft_ms": 56.85079097747803, "decode_ms": 124.23537496943027}, "total_wall_ms": 186.32574996445328, "ttft_ms": 56.85079097747803, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 379977728, "backend_memory": {"mlx_active_bytes": 6280634376, "mlx_peak_bytes": 6341282553, "mlx_cache_bytes": 3336182947, "process_rss_bytes": 379977728}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6341282553, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "\n\n", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "invalid_output", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7960b1fa79fbc02499ce6229a9a9eec92f86bde6fcca0fb9a9948b61cba15e6f", "output_hash": "75a11da44c802486bc6f65640aa48a730f0f684c5c07a42ba3cd1735eb3fb070"}149{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "506735b3474049ef8420f0fa", "attempt_id": "532b2e922f2f", "run_group_id": "E1_pilot:B2|bits=2|group_size=32", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:44.466503Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6262415368}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 2, "group_size": 32, "realized": {"bits": 2}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 13893120, "serialized_payload_bytes": 13926962, "side_information_bytes": 33842, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 70.37537498399615, "doc_prefill_ms": 978.4983330173418, "compress_ms": 56.51024996768683, "serialize_ms": 13.307290966622531, "load_ms": 3.502874984405935, "receiver_prefill_ms": 0.0, "query_prefill_ms": 208.85612501297146, "ttft_ms": 208.85612501297146, "decode_ms": 499.93570894002914}, "total_wall_ms": 714.4639589823782, "ttft_ms": 208.85612501297146, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 8}, "provider_cached_usage": null, "process_memory": 380092416, "backend_memory": {"mlx_active_bytes": 6280634376, "mlx_peak_bytes": 6336366736, "mlx_cache_bytes": 3336347395, "process_rss_bytes": 380092416}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6336366736, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "\n.\nenu0\n8.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7960b1fa79fbc02499ce6229a9a9eec92f86bde6fcca0fb9a9948b61cba15e6f", "output_hash": "0712057767b5231d8d43eb26fba1f7d44ff199739806ef7cef86faba42ba8c89"}150{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5cad66430628f529c1447990", "attempt_id": "25eef9c792e9", "run_group_id": "E1_pilot:B2|bits=2|group_size=32", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:45.206545Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6262415368}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 2, "group_size": 32, "realized": {"bits": 2}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 13893120, "serialized_payload_bytes": 13926962, "side_information_bytes": 33842, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 70.37537498399615, "doc_prefill_ms": 978.4983330173418, "compress_ms": 56.51024996768683, "serialize_ms": 13.307290966622531, "load_ms": 3.1956250313669443, "receiver_prefill_ms": 0.0, "query_prefill_ms": 82.27187499869615, "ttft_ms": 82.27187499869615, "decode_ms": 2853.5520839504898}, "total_wall_ms": 2941.789541975595, "ttft_ms": 82.27187499869615, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 380174336, "backend_memory": {"mlx_active_bytes": 6280634376, "mlx_peak_bytes": 6347167297, "mlx_cache_bytes": 3336372419, "process_rss_bytes": 380174336}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6347167297, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": " To()\ntoto\n\nto\ntoo\n\nto\n\nto\nto 11 10 0 1 0 0 0 0 000 ", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7960b1fa79fbc02499ce6229a9a9eec92f86bde6fcca0fb9a9948b61cba15e6f", "output_hash": "5c0fdcb95957f5817db3ee6259a246493e82fb3dbc571cdbe5b9e1fc64507105"}151{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "45449d4dadf2ca7f494fb1c1", "attempt_id": "cf7d26ec9c79", "run_group_id": "E1_pilot:B3|keep_frac=0.5", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:48.276243Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6285123592}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.5, "realized": {"policy": "sink_recent", "keep_frac": 0.5, "T": 2010, "keep": 1005, "compress_ms": 80.88529098313302}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064494, "side_information_bytes": 16174, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 103.52700005751103, "doc_prefill_ms": 978.4983330173418, "compress_ms": 80.96541708800942, "serialize_ms": 22.373708081431687, "load_ms": 1.7549169715493917, "receiver_prefill_ms": 0.0, "query_prefill_ms": 66.08933303505182, "ttft_ms": 66.08933303505182, "decode_ms": 104.69000006560236}, "total_wall_ms": 174.91079203318805, "ttft_ms": 66.08933303505182, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 391757824, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6363029708, "mlx_cache_bytes": 3292954819, "process_rss_bytes": 391757824}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6363029708, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Not provided", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "17d15600096a15c4e1adfeb19b160a6cab5f5501a89db769a2c4e0577f8edbc5", "output_hash": "d83b9ff3c07a8390e2ea0a2af45fba37595da25f31b85ba625363ee98c0c796f"}152{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "874a877b8fc70e6daa01ddbe", "attempt_id": "aa03ce999f06", "run_group_id": "E1_pilot:B3|keep_frac=0.5", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:48.475586Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6285123592}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.5, "realized": {"policy": "sink_recent", "keep_frac": 0.5, "T": 2010, "keep": 1005, "compress_ms": 80.88529098313302}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064494, "side_information_bytes": 16174, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 103.52700005751103, "doc_prefill_ms": 978.4983330173418, "compress_ms": 80.96541708800942, "serialize_ms": 22.373708081431687, "load_ms": 1.1513339122757316, "receiver_prefill_ms": 0.0, "query_prefill_ms": 65.51987503189594, "ttft_ms": 65.51987503189594, "decode_ms": 501.2499999720603}, "total_wall_ms": 569.7316660080105, "ttft_ms": 65.51987503189594, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 391757824, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6369239268, "mlx_cache_bytes": 3292954819, "process_rss_bytes": 391757824}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6369239268, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "17d15600096a15c4e1adfeb19b160a6cab5f5501a89db769a2c4e0577f8edbc5", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}153{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0bd9a0f0bf57a1aebfe6b976", "attempt_id": "5cd68886b63f", "run_group_id": "E1_pilot:B3|keep_frac=0.5", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:49.070240Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6285123592}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.5, "realized": {"policy": "sink_recent", "keep_frac": 0.5, "T": 2010, "keep": 1005, "compress_ms": 80.88529098313302}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064494, "side_information_bytes": 16174, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 103.52700005751103, "doc_prefill_ms": 978.4983330173418, "compress_ms": 80.96541708800942, "serialize_ms": 22.373708081431687, "load_ms": 1.1686249636113644, "receiver_prefill_ms": 0.0, "query_prefill_ms": 51.905416999943554, "ttft_ms": 51.905416999943554, "decode_ms": 385.2078339550644}, "total_wall_ms": 440.02754206303507, "ttft_ms": 51.905416999943554, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 391790592, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6369239268, "mlx_cache_bytes": 3292954819, "process_rss_bytes": 391790592}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6369239268, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "17d15600096a15c4e1adfeb19b160a6cab5f5501a89db769a2c4e0577f8edbc5", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}154{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8ac4718d4962b7af41d1be8e", "attempt_id": "ec2f80111d8b", "run_group_id": "E1_pilot:B3|keep_frac=0.5", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:49.534245Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6285123592}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.5, "realized": {"policy": "sink_recent", "keep_frac": 0.5, "T": 2010, "keep": 1005, "compress_ms": 80.88529098313302}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064494, "side_information_bytes": 16174, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 103.52700005751103, "doc_prefill_ms": 978.4983330173418, "compress_ms": 80.96541708800942, "serialize_ms": 22.373708081431687, "load_ms": 1.8943339819088578, "receiver_prefill_ms": 0.0, "query_prefill_ms": 73.95041699055582, "ttft_ms": 73.95041699055582, "decode_ms": 264.9809579597786}, "total_wall_ms": 342.73837495129555, "ttft_ms": 73.95041699055582, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 391790592, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6368223460, "mlx_cache_bytes": 3292954819, "process_rss_bytes": 391790592}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6368223460, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "17d15600096a15c4e1adfeb19b160a6cab5f5501a89db769a2c4e0577f8edbc5", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}155{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "4b8380cb52bc8ff97b46c252", "attempt_id": "da668159a6fa", "run_group_id": "E1_pilot:B3|keep_frac=0.5", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:49.906444Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6285123592}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.5, "realized": {"policy": "sink_recent", "keep_frac": 0.5, "T": 2010, "keep": 1005, "compress_ms": 80.88529098313302}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064494, "side_information_bytes": 16174, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 103.52700005751103, "doc_prefill_ms": 978.4983330173418, "compress_ms": 80.96541708800942, "serialize_ms": 22.373708081431687, "load_ms": 6.182792014442384, "receiver_prefill_ms": 0.0, "query_prefill_ms": 64.67024993617088, "ttft_ms": 64.67024993617088, "decode_ms": 362.70954203791916}, "total_wall_ms": 435.4581250809133, "ttft_ms": 64.67024993617088, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 391872512, "backend_memory": {"mlx_active_bytes": 6325231624, "mlx_peak_bytes": 6372630776, "mlx_cache_bytes": 3341910211, "process_rss_bytes": 391872512}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6372630776, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "7354655", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "17d15600096a15c4e1adfeb19b160a6cab5f5501a89db769a2c4e0577f8edbc5", "output_hash": "8d148a11998d8f65810559f1ef4ae6479852ba925e986cd71d6da7e4a8e315fb"}156{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "1e0880ce4166762fa73a7056", "attempt_id": "16f7dd0ee112", "run_group_id": "E1_pilot:B3|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:50.431274Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 47.57491697091609}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521645, "side_information_bytes": 15917, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 60.686292010359466, "doc_prefill_ms": 978.4983330173418, "compress_ms": 47.63674992136657, "serialize_ms": 12.81133305747062, "load_ms": 1.2637079926207662, "receiver_prefill_ms": 0.0, "query_prefill_ms": 78.25625000987202, "ttft_ms": 78.25625000987202, "decode_ms": 410.6784160248935}, "total_wall_ms": 491.99295800644904, "ttft_ms": 78.25625000987202, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 415449088, "backend_memory": {"mlx_active_bytes": 6287548424, "mlx_peak_bytes": 6321594572, "mlx_cache_bytes": 3406203035, "process_rss_bytes": 415449088}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6321594572, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "MP-89X-6320", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f6ef81007573cfaec410eb9847c70ee516bcb552b3345400893355ca501b96b", "output_hash": "e07259f8346a7570e5b980e01a868e6af8ede517e6d7f6a2958e2f3b8da08023"}157{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "53d8f50534dd727bbd148b91", "attempt_id": "054cfe16836b", "run_group_id": "E1_pilot:B3|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:50.947421Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 47.57491697091609}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521645, "side_information_bytes": 15917, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 60.686292010359466, "doc_prefill_ms": 978.4983330173418, "compress_ms": 47.63674992136657, "serialize_ms": 12.81133305747062, "load_ms": 1.2311659520491958, "receiver_prefill_ms": 0.0, "query_prefill_ms": 31.247791019268334, "ttft_ms": 31.247791019268334, "decode_ms": 361.6451659472659}, "total_wall_ms": 396.09987498261034, "ttft_ms": 31.247791019268334, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 415481856, "backend_memory": {"mlx_active_bytes": 6287548424, "mlx_peak_bytes": 6327804132, "mlx_cache_bytes": 3406203035, "process_rss_bytes": 415481856}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327804132, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f6ef81007573cfaec410eb9847c70ee516bcb552b3345400893355ca501b96b", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}158{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "63c86e53f1a9da4ee1453a9d", "attempt_id": "d2bbbec71c5c", "run_group_id": "E1_pilot:B3|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:51.367094Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 47.57491697091609}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521645, "side_information_bytes": 15917, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 60.686292010359466, "doc_prefill_ms": 978.4983330173418, "compress_ms": 47.63674992136657, "serialize_ms": 12.81133305747062, "load_ms": 2.270334050990641, "receiver_prefill_ms": 0.0, "query_prefill_ms": 54.82562503311783, "ttft_ms": 54.82562503311783, "decode_ms": 98.71366701554507}, "total_wall_ms": 157.672708039172, "ttft_ms": 54.82562503311783, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 415498240, "backend_memory": {"mlx_active_bytes": 6287482888, "mlx_peak_bytes": 6327804132, "mlx_cache_bytes": 3406268571, "process_rss_bytes": 415498240}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327804132, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Not provided", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f6ef81007573cfaec410eb9847c70ee516bcb552b3345400893355ca501b96b", "output_hash": "d83b9ff3c07a8390e2ea0a2af45fba37595da25f31b85ba625363ee98c0c796f"}159{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2f15871722d1aef492ba94a7", "attempt_id": "3b1765607a66", "run_group_id": "E1_pilot:B3|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:51.551640Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 47.57491697091609}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521645, "side_information_bytes": 15917, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 60.686292010359466, "doc_prefill_ms": 978.4983330173418, "compress_ms": 47.63674992136657, "serialize_ms": 12.81133305747062, "load_ms": 1.915832981467247, "receiver_prefill_ms": 0.0, "query_prefill_ms": 52.56808397825807, "ttft_ms": 52.56808397825807, "decode_ms": 114.58470893558115}, "total_wall_ms": 171.09187494497746, "ttft_ms": 52.56808397825807, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 415531008, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6326788324, "mlx_cache_bytes": 3407448219, "process_rss_bytes": 415531008}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6326788324, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Milos Ferrante", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f6ef81007573cfaec410eb9847c70ee516bcb552b3345400893355ca501b96b", "output_hash": "14ae3adaefbe5d307a9585834a37951eb0794dc1bfacd0739a580886526143d2"}160{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "7b5d7a113122635eb04bccbe", "attempt_id": "86520a116129", "run_group_id": "E1_pilot:B3|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:51.746736Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 47.57491697091609}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521645, "side_information_bytes": 15917, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 60.686292010359466, "doc_prefill_ms": 978.4983330173418, "compress_ms": 47.63674992136657, "serialize_ms": 12.81133305747062, "load_ms": 1.3356669805943966, "receiver_prefill_ms": 0.0, "query_prefill_ms": 51.01012496743351, "ttft_ms": 51.01012496743351, "decode_ms": 254.03487496078014}, "total_wall_ms": 308.31062502693385, "ttft_ms": 51.01012496743351, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 6}, "provider_cached_usage": null, "process_memory": 415531008, "backend_memory": {"mlx_active_bytes": 6287548424, "mlx_peak_bytes": 6332375288, "mlx_cache_bytes": 3406203035, "process_rss_bytes": 415531008}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6332375288, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "120000", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f6ef81007573cfaec410eb9847c70ee516bcb552b3345400893355ca501b96b", "output_hash": "4f9f73b34c5b89879aad65a48025f3187dd9ce6dc3d4e88eecb2fc79227350f1"}161{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "def178164103808ffb24b221", "attempt_id": "0fdd6ec712c9", "run_group_id": "E1_pilot:B3|keep_frac=0.125", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:52.129712Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6256812040}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.125, "realized": {"policy": "sink_recent", "keep_frac": 0.125, "T": 2010, "keep": 251, "compress_ms": 41.31716606207192}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268708, "side_information_bytes": 15844, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 47.348666936159134, "doc_prefill_ms": 978.4983330173418, "compress_ms": 41.36320808902383, "serialize_ms": 5.754875019192696, "load_ms": 1.6183329280465841, "receiver_prefill_ms": 0.0, "query_prefill_ms": 48.33358305040747, "ttft_ms": 48.33358305040747, "decode_ms": 421.7722500907257}, "total_wall_ms": 473.82229194045067, "ttft_ms": 48.33358305040747, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 419807232, "backend_memory": {"mlx_active_bytes": 6269329416, "mlx_peak_bytes": 6301605788, "mlx_cache_bytes": 3424749723, "process_rss_bytes": 419807232}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6301605788, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "MP-89X-6320", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "55f4e09494fa3cefd3b5233346303ec551b5e3340638f269e7a42ddf548197cf", "output_hash": "e07259f8346a7570e5b980e01a868e6af8ede517e6d7f6a2958e2f3b8da08023"}162{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ce78979af3b845da8fc9d33a", "attempt_id": "7689c26ad116", "run_group_id": "E1_pilot:B3|keep_frac=0.125", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:52.628211Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6256812040}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.125, "realized": {"policy": "sink_recent", "keep_frac": 0.125, "T": 2010, "keep": 251, "compress_ms": 41.31716606207192}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268708, "side_information_bytes": 15844, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 47.348666936159134, "doc_prefill_ms": 978.4983330173418, "compress_ms": 41.36320808902383, "serialize_ms": 5.754875019192696, "load_ms": 1.2893329840153456, "receiver_prefill_ms": 0.0, "query_prefill_ms": 37.5941249076277, "ttft_ms": 37.5941249076277, "decode_ms": 300.26895890478045}, "total_wall_ms": 340.8969999291003, "ttft_ms": 37.5941249076277, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 419840000, "backend_memory": {"mlx_active_bytes": 6269296648, "mlx_peak_bytes": 6309404596, "mlx_cache_bytes": 3428059291, "process_rss_bytes": 419840000}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6309404596, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "55f4e09494fa3cefd3b5233346303ec551b5e3340638f269e7a42ddf548197cf", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}163{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0198af13b1129c2e71e5a668", "attempt_id": "6eb56a3cac72", "run_group_id": "E1_pilot:B3|keep_frac=0.125", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:52.992584Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6256812040}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.125, "realized": {"policy": "sink_recent", "keep_frac": 0.125, "T": 2010, "keep": 251, "compress_ms": 41.31716606207192}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268708, "side_information_bytes": 15844, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 47.348666936159134, "doc_prefill_ms": 978.4983330173418, "compress_ms": 41.36320808902383, "serialize_ms": 5.754875019192696, "load_ms": 1.2312500039115548, "receiver_prefill_ms": 0.0, "query_prefill_ms": 47.74345795158297, "ttft_ms": 47.74345795158297, "decode_ms": 84.82758305035532}, "total_wall_ms": 135.71087492164224, "ttft_ms": 47.74345795158297, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 419840000, "backend_memory": {"mlx_active_bytes": 6269083656, "mlx_peak_bytes": 6308945844, "mlx_cache_bytes": 3429746843, "process_rss_bytes": 419840000}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6308945844, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Not provided", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "55f4e09494fa3cefd3b5233346303ec551b5e3340638f269e7a42ddf548197cf", "output_hash": "d83b9ff3c07a8390e2ea0a2af45fba37595da25f31b85ba625363ee98c0c796f"}164{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8e7267c6b703784866e0b8ef", "attempt_id": "7ed61cb6de64", "run_group_id": "E1_pilot:B3|keep_frac=0.125", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:53.153119Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6256812040}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.125, "realized": {"policy": "sink_recent", "keep_frac": 0.125, "T": 2010, "keep": 251, "compress_ms": 41.31716606207192}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268708, "side_information_bytes": 15844, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 47.348666936159134, "doc_prefill_ms": 978.4983330173418, "compress_ms": 41.36320808902383, "serialize_ms": 5.754875019192696, "load_ms": 2.68454197794199, "receiver_prefill_ms": 0.0, "query_prefill_ms": 42.02141589485109, "ttft_ms": 42.02141589485109, "decode_ms": 96.01737488992512}, "total_wall_ms": 142.68562500365078, "ttft_ms": 42.02141589485109, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 419889152, "backend_memory": {"mlx_active_bytes": 6268739592, "mlx_peak_bytes": 6306291636, "mlx_cache_bytes": 3433367707, "process_rss_bytes": 419889152}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6306291636, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "John Doe", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "55f4e09494fa3cefd3b5233346303ec551b5e3340638f269e7a42ddf548197cf", "output_hash": "6cea57c2fb6cbc2a40411135005760f241fffc3e5e67ab99882726431037f908"}165{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "06d010949bc01ca19ade1895", "attempt_id": "8220c61d5c46", "run_group_id": "E1_pilot:B3|keep_frac=0.125", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:53.321397Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6256812040}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.125, "realized": {"policy": "sink_recent", "keep_frac": 0.125, "T": 2010, "keep": 251, "compress_ms": 41.31716606207192}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268708, "side_information_bytes": 15844, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 47.348666936159134, "doc_prefill_ms": 978.4983330173418, "compress_ms": 41.36320808902383, "serialize_ms": 5.754875019192696, "load_ms": 1.3275420060381293, "receiver_prefill_ms": 0.0, "query_prefill_ms": 48.03954192902893, "ttft_ms": 48.03954192902893, "decode_ms": 218.25925004668534}, "total_wall_ms": 269.2964170128107, "ttft_ms": 48.03954192902893, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 6}, "provider_cached_usage": null, "process_memory": 419889152, "backend_memory": {"mlx_active_bytes": 6268821512, "mlx_peak_bytes": 6311649224, "mlx_cache_bytes": 3433285787, "process_rss_bytes": 419889152}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6311649224, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "500000", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "55f4e09494fa3cefd3b5233346303ec551b5e3340638f269e7a42ddf548197cf", "output_hash": "8d6962a152aee235ba824c41758b8da2371b7077b4ea0afaaec94014e16e3bc7"}166{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "05a7b029b6a4058f6b259ccc", "attempt_id": "a05fc52f44de", "run_group_id": "E1_pilot:B4|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:53.682878Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B4", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "random", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 57.059750077314675}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521641, "side_information_bytes": 15913, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 68.19895806256682, "doc_prefill_ms": 978.4983330173418, "compress_ms": 57.12962499819696, "serialize_ms": 10.83554106298834, "load_ms": 1.6067089745774865, "receiver_prefill_ms": 0.0, "query_prefill_ms": 40.13862495776266, "ttft_ms": 40.13862495776266, "decode_ms": 1860.7685409951955}, "total_wall_ms": 1904.3388329446316, "ttft_ms": 40.13862495776266, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 435650560, "backend_memory": {"mlx_active_bytes": 6288662536, "mlx_peak_bytes": 6321594572, "mlx_cache_bytes": 3439593627, "process_rss_bytes": 435650560}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6321594572, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "assistant\n\nAssistant, Assistant, Assistant\n\nAssistant, Assistant, Assistant\n\nAssistant, Assistant, Assistant\n\nAssistant, Assistant, Assistant\n\nAssistant, Assistant, Assistant\n\nAssistant, Assistant, Assistant\n\nAssistant, Assistant, Assistant\n\nAssistant, Assistant,", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "febe22ff4556ec505bab21f2ebddecefa50d28e8b3bba2fcc597978e4f400e24", "output_hash": "a7f3f386f2ebe03f3097faba3f505d405d26b2f9abd5b6faacedc39b50901a2d"}167{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "52ac8dc34857e9b2519ed43e", "attempt_id": "531c6105b4f0", "run_group_id": "E1_pilot:B4|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:55.609700Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B4", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "random", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 57.059750077314675}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521641, "side_information_bytes": 15913, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 68.19895806256682, "doc_prefill_ms": 978.4983330173418, "compress_ms": 57.12962499819696, "serialize_ms": 10.83554106298834, "load_ms": 2.5329169584438205, "receiver_prefill_ms": 0.0, "query_prefill_ms": 52.1832499653101, "ttft_ms": 52.1832499653101, "decode_ms": 1838.2992920232937}, "total_wall_ms": 1894.7905829409137, "ttft_ms": 52.1832499653101, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 435683328, "backend_memory": {"mlx_active_bytes": 6288662536, "mlx_peak_bytes": 6327804132, "mlx_cache_bytes": 3439593627, "process_rss_bytes": 435683328}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327804132, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "I am a memo. I am a memo. I am a memo. I am a memo. I am a memo. I am a memo. I am a memo. I am a memo. I am a memo. I am a", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "febe22ff4556ec505bab21f2ebddecefa50d28e8b3bba2fcc597978e4f400e24", "output_hash": "426a0f227def734f39a983ae88dcd05e8dcbe992627914324fb637eb8cb63e6b"}168{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "333c5a7b7e0a8e5455a49847", "attempt_id": "be34f0389e53", "run_group_id": "E1_pilot:B4|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:57.527494Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B4", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "random", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 57.059750077314675}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521641, "side_information_bytes": 15913, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 68.19895806256682, "doc_prefill_ms": 978.4983330173418, "compress_ms": 57.12962499819696, "serialize_ms": 10.83554106298834, "load_ms": 3.0611660331487656, "receiver_prefill_ms": 0.0, "query_prefill_ms": 52.79670807067305, "ttft_ms": 52.79670807067305, "decode_ms": 1919.1165410447866}, "total_wall_ms": 1977.272915886715, "ttft_ms": 52.79670807067305, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 435716096, "backend_memory": {"mlx_active_bytes": 6288662536, "mlx_peak_bytes": 6327804132, "mlx_cache_bytes": 3439593627, "process_rss_bytes": 435716096}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327804132, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "The Project codes are the name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name name", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "febe22ff4556ec505bab21f2ebddecefa50d28e8b3bba2fcc597978e4f400e24", "output_hash": "79576cc33fa68a3208636b2d8810c8c0d1f2117032776dc346a524e04fb85a25"}169{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8bac4567a99d3b11349b3c4d", "attempt_id": "b42e3cb96243", "run_group_id": "E1_pilot:B4|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T03:59:59.530821Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B4", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "random", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 57.059750077314675}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521641, "side_information_bytes": 15913, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 68.19895806256682, "doc_prefill_ms": 978.4983330173418, "compress_ms": 57.12962499819696, "serialize_ms": 10.83554106298834, "load_ms": 2.693540998734534, "receiver_prefill_ms": 0.0, "query_prefill_ms": 44.54070806968957, "ttft_ms": 44.54070806968957, "decode_ms": 1905.63195804134}, "total_wall_ms": 1955.128541099839, "ttft_ms": 44.54070806968957, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 435765248, "backend_memory": {"mlx_active_bytes": 6288662536, "mlx_peak_bytes": 6324396260, "mlx_cache_bytes": 3439593627, "process_rss_bytes": 435765248}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6324396260, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "I will revise the project.\nI will revise the project.\nI will revise the project.\nI will revise the project.\nI will revise the project.\nI will revise the project.\nI will revise the project.\nI will revise the project.\n", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "febe22ff4556ec505bab21f2ebddecefa50d28e8b3bba2fcc597978e4f400e24", "output_hash": "ec85516726e09bae04746f5b6be22f47099830d26c5193e069713a8e4b00dca5"}170{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "40ce8ee2cc459e53d432ac9b", "attempt_id": "c31b11fcc433", "run_group_id": "E1_pilot:B4|keep_frac=0.25", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:01.509515Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B4", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "random", "keep_frac": 0.25, "T": 2010, "keep": 502, "compress_ms": 57.059750077314675}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521641, "side_information_bytes": 15913, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 68.19895806256682, "doc_prefill_ms": 978.4983330173418, "compress_ms": 57.12962499819696, "serialize_ms": 10.83554106298834, "load_ms": 1.6787908971309662, "receiver_prefill_ms": 0.0, "query_prefill_ms": 35.12537502683699, "ttft_ms": 35.12537502683699, "decode_ms": 1861.2989590037614}, "total_wall_ms": 1900.1809590263292, "ttft_ms": 35.12537502683699, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 435765248, "backend_memory": {"mlx_active_bytes": 6288662536, "mlx_peak_bytes": 6332375288, "mlx_cache_bytes": 3439593627, "process_rss_bytes": 435765248}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6332375288, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "The Project Torsel-111111111111111111111111111111111111111111", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "febe22ff4556ec505bab21f2ebddecefa50d28e8b3bba2fcc597978e4f400e24", "output_hash": "507026f6b067a372cc62e223c8f35d069220ce4324d41766c0e2fb2fee2e3554"}171{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3303177bc121ddf9c61342b0", "attempt_id": "d2bf00d69e03", "run_group_id": "E1_pilot:B5|keep_frac=0.5|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:03.435298Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.5, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.5, "T": 2010, "keep": 1005, "window_tokens": 35, "compress_ms": 131.71937502920628}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064535, "side_information_bytes": 16215, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 149.07495898660272, "doc_prefill_ms": 978.4983330173418, "compress_ms": 134.04241693206131, "serialize_ms": 14.688875060528517, "load_ms": 2.247416996397078, "receiver_prefill_ms": 0.0, "query_prefill_ms": 63.857041066512465, "ttft_ms": 63.857041066512465, "decode_ms": 163.83012500591576}, "total_wall_ms": 385.94475004356354, "ttft_ms": 63.857041066512465, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 450183168, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6363029708, "mlx_cache_bytes": 3493365915, "process_rss_bytes": 450183168}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6363029708, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "91f86923cebcc9c4b68998c5f39e5039c9052204124a0962d5e4b3d2125e9404", "output_hash": "4507f73e53dc60cf2db0c66127c210e76c026e3a3f2cb327b372cd61793443d9"}172{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2f4b217f1ef51bfcb90f707f", "attempt_id": "8a6a4ada194b", "run_group_id": "E1_pilot:B5|keep_frac=0.5|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:03.850169Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.5, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.5, "T": 2010, "keep": 1005, "window_tokens": 41, "compress_ms": 135.56512503419071}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064535, "side_information_bytes": 16215, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 152.8303329832852, "doc_prefill_ms": 978.4983330173418, "compress_ms": 137.6165000256151, "serialize_ms": 14.900958980433643, "load_ms": 0.5778749473392963, "receiver_prefill_ms": 0.0, "query_prefill_ms": 38.181290961802006, "ttft_ms": 38.181290961802006, "decode_ms": 373.5233750194311}, "total_wall_ms": 567.1716249780729, "ttft_ms": 38.181290961802006, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 459767808, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6368649444, "mlx_cache_bytes": 3502049435, "process_rss_bytes": 459767808}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6368649444, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "91f86923cebcc9c4b68998c5f39e5039c9052204124a0962d5e4b3d2125e9404", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}173{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c876c85a9f2b61a3c8031170", "attempt_id": "b4113e430ef9", "run_group_id": "E1_pilot:B5|keep_frac=0.5|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:04.446646Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.5, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.5, "T": 2010, "keep": 1005, "window_tokens": 41, "compress_ms": 181.6300000064075}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064534, "side_information_bytes": 16214, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 197.6571249542758, "doc_prefill_ms": 978.4983330173418, "compress_ms": 183.52724995929748, "serialize_ms": 13.731124927289784, "load_ms": 1.8607500242069364, "receiver_prefill_ms": 0.0, "query_prefill_ms": 55.253040976822376, "ttft_ms": 55.253040976822376, "decode_ms": 255.760750034824}, "total_wall_ms": 513.2913329871371, "ttft_ms": 55.253040976822376, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 478691328, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6368649444, "mlx_cache_bytes": 3502049435, "process_rss_bytes": 478691328}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6368649444, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4820105fb795b0f6f33d2f15ee475ab2efa4add51302a6e39353817db749e573", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}174{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "03dc480f0976181173edc584", "attempt_id": "df26328f822d", "run_group_id": "E1_pilot:B5|keep_frac=0.5|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:04.987402Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.5, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.5, "T": 2010, "keep": 1005, "window_tokens": 38, "compress_ms": 158.82058395072818}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064535, "side_information_bytes": 16215, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 173.08904195670038, "doc_prefill_ms": 978.4983330173418, "compress_ms": 160.90166696812958, "serialize_ms": 11.821750085800886, "load_ms": 0.841250061057508, "receiver_prefill_ms": 0.0, "query_prefill_ms": 59.6971670165658, "ttft_ms": 59.6971670165658, "decode_ms": 168.02791599184275}, "total_wall_ms": 403.6270839860663, "ttft_ms": 59.6971670165658, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 478822400, "backend_memory": {"mlx_active_bytes": 6324051976, "mlx_peak_bytes": 6365831396, "mlx_cache_bytes": 3506948251, "process_rss_bytes": 478822400}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6365831396, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "91f86923cebcc9c4b68998c5f39e5039c9052204124a0962d5e4b3d2125e9404", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}175{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "111fbedca997903c677bda6d", "attempt_id": "ac76d9517767", "run_group_id": "E1_pilot:B5|keep_frac=0.5|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:05.419802Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.5, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.5, "T": 2010, "keep": 1005, "window_tokens": 46, "compress_ms": 174.6552080148831}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 37048320, "serialized_payload_bytes": 37064534, "side_information_bytes": 16214, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 188.59779089689255, "doc_prefill_ms": 978.4983330173418, "compress_ms": 176.50412500370294, "serialize_ms": 11.79779099766165, "load_ms": 2.6412499137222767, "receiver_prefill_ms": 0.0, "query_prefill_ms": 44.651041040197015, "ttft_ms": 44.651041040197015, "decode_ms": 262.3440419556573}, "total_wall_ms": 500.2535410458222, "ttft_ms": 44.651041040197015, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 6}, "provider_cached_usage": null, "process_memory": 479117312, "backend_memory": {"mlx_active_bytes": 6325231624, "mlx_peak_bytes": 6372630776, "mlx_cache_bytes": 3526445211, "process_rss_bytes": 479117312}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6372630776, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "688172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4820105fb795b0f6f33d2f15ee475ab2efa4add51302a6e39353817db749e573", "output_hash": "dfd43f8ddbe767bad909ada1437a79c0dbd36d9b33121c8a65a6b3eebf334586"}176{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "80a96a5619533c79772081f4", "attempt_id": "156856afeea5", "run_group_id": "E1_pilot:B5|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:05.944797Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 35, "compress_ms": 143.22600001469254}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521686, "side_information_bytes": 15958, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 153.58562499750406, "doc_prefill_ms": 978.4983330173418, "compress_ms": 146.09579101670533, "serialize_ms": 7.035665912553668, "load_ms": 1.6515000024810433, "receiver_prefill_ms": 0.0, "query_prefill_ms": 45.96708307508379, "ttft_ms": 45.96708307508379, "decode_ms": 276.07845899183303}, "total_wall_ms": 479.32941594626755, "ttft_ms": 45.96708307508379, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 6}, "provider_cached_usage": null, "process_memory": 480608256, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6321594572, "mlx_cache_bytes": 3565373595, "process_rss_bytes": 480608256}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6321594572, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-1482", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2133e2d8fb6fbe8bb2bc071e98a75ef68a33983b8f86299196b1d6fddcd4bf21", "output_hash": "f5e9a6d78321ae31e0a3bdeeaf8c39b796c1fc12a0a8896955e8c411f428367c"}177{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9a66b7b6a85827a424502a1d", "attempt_id": "f95e1bf5622b", "run_group_id": "E1_pilot:B5|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:06.456285Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 103.93987502902746}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521686, "side_information_bytes": 15958, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 113.22887509595603, "doc_prefill_ms": 978.4983330173418, "compress_ms": 105.88241706136614, "serialize_ms": 7.026750012300909, "load_ms": 0.954542076215148, "receiver_prefill_ms": 0.0, "query_prefill_ms": 37.40387503057718, "ttft_ms": 37.40387503057718, "decode_ms": 384.2731670010835}, "total_wall_ms": 538.0691250320524, "ttft_ms": 37.40387503057718, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 480641024, "backend_memory": {"mlx_active_bytes": 6287548424, "mlx_peak_bytes": 6327214308, "mlx_cache_bytes": 3564128411, "process_rss_bytes": 480641024}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327214308, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2133e2d8fb6fbe8bb2bc071e98a75ef68a33983b8f86299196b1d6fddcd4bf21", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}178{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "e43a496a2698ce861355b218", "attempt_id": "b5655d8b8347", "run_group_id": "E1_pilot:B5|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:07.023641Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 133.6641670204699}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521685, "side_information_bytes": 15957, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 141.8685419484973, "doc_prefill_ms": 978.4983330173418, "compress_ms": 135.57112507987767, "serialize_ms": 5.908999941311777, "load_ms": 0.8796249749138951, "receiver_prefill_ms": 0.0, "query_prefill_ms": 31.09145793132484, "ttft_ms": 31.09145793132484, "decode_ms": 301.63462506607175}, "total_wall_ms": 477.33141598291695, "ttft_ms": 31.09145793132484, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 480641024, "backend_memory": {"mlx_active_bytes": 6287482888, "mlx_peak_bytes": 6327214308, "mlx_cache_bytes": 3564193947, "process_rss_bytes": 480641024}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327214308, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c45b4f30b880b0f5227d71d6f0c6992840fbb19374bbf7cfe2bdd05e112c7db5", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}179{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "998f9b86ffb98e798bf6d872", "attempt_id": "7aa253603792", "run_group_id": "E1_pilot:B5|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:07.527466Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 38, "compress_ms": 187.73520807735622}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521686, "side_information_bytes": 15958, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 202.1150829968974, "doc_prefill_ms": 978.4983330173418, "compress_ms": 189.45208296645433, "serialize_ms": 12.313249986618757, "load_ms": 1.4884580159559846, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.767624955624342, "ttft_ms": 20.767624955624342, "decode_ms": 187.98804108519107}, "total_wall_ms": 418.6952089658007, "ttft_ms": 20.767624955624342, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 481083392, "backend_memory": {"mlx_active_bytes": 6287482888, "mlx_peak_bytes": 6324396260, "mlx_cache_bytes": 3564193947, "process_rss_bytes": 481083392}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6324396260, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2133e2d8fb6fbe8bb2bc071e98a75ef68a33983b8f86299196b1d6fddcd4bf21", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}180{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c0019074c06e979f48a16ca5", "attempt_id": "5304d2e9139a", "run_group_id": "E1_pilot:B5|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:07.970954Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 46, "compress_ms": 170.56350002530962}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521686, "side_information_bytes": 15958, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 177.93058301322162, "doc_prefill_ms": 978.4983330173418, "compress_ms": 172.0557090593502, "serialize_ms": 5.5994169088080525, "load_ms": 1.487958012148738, "receiver_prefill_ms": 0.0, "query_prefill_ms": 58.72408300638199, "ttft_ms": 58.72408300638199, "decode_ms": 276.4550409046933}, "total_wall_ms": 516.4899170631543, "ttft_ms": 58.72408300638199, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 481116160, "backend_memory": {"mlx_active_bytes": 6287548424, "mlx_peak_bytes": 6332375288, "mlx_cache_bytes": 3564128411, "process_rss_bytes": 481116160}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6332375288, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2133e2d8fb6fbe8bb2bc071e98a75ef68a33983b8f86299196b1d6fddcd4bf21", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}181{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "647b64c583db2c5cadd31c55", "attempt_id": "5196d54eaa5a", "run_group_id": "E1_pilot:B5|keep_frac=0.125|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:08.510922Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.125, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.125, "T": 2010, "keep": 251, "window_tokens": 35, "compress_ms": 162.36558405216783}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268749, "side_information_bytes": 15885, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 169.9978749966249, "doc_prefill_ms": 978.4983330173418, "compress_ms": 164.06458406709135, "serialize_ms": 5.610708030872047, "load_ms": 0.975457951426506, "receiver_prefill_ms": 0.0, "query_prefill_ms": 52.04462504480034, "ttft_ms": 52.04462504480034, "decode_ms": 219.67383404262364}, "total_wall_ms": 445.0735829304904, "ttft_ms": 52.04462504480034, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 482000896, "backend_memory": {"mlx_active_bytes": 6268821512, "mlx_peak_bytes": 6301605788, "mlx_cache_bytes": 3582855323, "process_rss_bytes": 482000896}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6301605788, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-141", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c3d5da842f16f311212d8a70f4b9b00acfc49bad8c8a6407cc8219c2e896e52d", "output_hash": "a9b587de3507f5e878743054a9bd92f5ac43e5c3f7d1892fa950800e78d954cf"}182{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6450f6180643834c1f28063f", "attempt_id": "4aca836749b6", "run_group_id": "E1_pilot:B5|keep_frac=0.125|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:08.979331Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.125, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.125, "T": 2010, "keep": 251, "window_tokens": 41, "compress_ms": 158.90608401969075}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268749, "side_information_bytes": 15885, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 166.8117920635268, "doc_prefill_ms": 978.4983330173418, "compress_ms": 160.8652089489624, "serialize_ms": 5.643125041387975, "load_ms": 3.785416018217802, "receiver_prefill_ms": 0.0, "query_prefill_ms": 46.947708004154265, "ttft_ms": 46.947708004154265, "decode_ms": 366.770250024274}, "total_wall_ms": 586.5151250036433, "ttft_ms": 46.947708004154265, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 8}, "provider_cached_usage": null, "process_memory": 484786176, "backend_memory": {"mlx_active_bytes": 6268821512, "mlx_peak_bytes": 6307667892, "mlx_cache_bytes": 3582855323, "process_rss_bytes": 484786176}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6307667892, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.5714285714285715}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c3d5da842f16f311212d8a70f4b9b00acfc49bad8c8a6407cc8219c2e896e52d", "output_hash": "2b9f266644d15d5e9fa69e24c25c5efa240ca43aa71c2a39e46dfa6ce23ba0b2"}183{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "4e99942e56b141649e7418ca", "attempt_id": "e677b431959e", "run_group_id": "E1_pilot:B5|keep_frac=0.125|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:09.592901Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.125, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.125, "T": 2010, "keep": 251, "window_tokens": 41, "compress_ms": 91.04545903392136}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268748, "side_information_bytes": 15884, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 97.79495897237211, "doc_prefill_ms": 978.4983330173418, "compress_ms": 92.75412489660084, "serialize_ms": 4.724457976408303, "load_ms": 0.4404169740155339, "receiver_prefill_ms": 0.0, "query_prefill_ms": 45.32216698862612, "ttft_ms": 45.32216698862612, "decode_ms": 285.37225001491606}, "total_wall_ms": 430.65704102627933, "ttft_ms": 45.32216698862612, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 484835328, "backend_memory": {"mlx_active_bytes": 6268821512, "mlx_peak_bytes": 6307667892, "mlx_cache_bytes": 3582855323, "process_rss_bytes": 484835328}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6307667892, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "a6be719c0cf9eeedd4af5921cc6f6ec7b9168b4dc130f226b184594994e2cbcc", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}184{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5594d171faf427afef1e9cfd", "attempt_id": "0a3bbeda7871", "run_group_id": "E1_pilot:B5|keep_frac=0.125|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:10.047096Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.125, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.125, "T": 2010, "keep": 251, "window_tokens": 38, "compress_ms": 122.17254098504782}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268749, "side_information_bytes": 15885, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 128.34070902317762, "doc_prefill_ms": 978.4983330173418, "compress_ms": 123.67729190737009, "serialize_ms": 4.361208062618971, "load_ms": 1.535917050205171, "receiver_prefill_ms": 0.0, "query_prefill_ms": 60.15229190234095, "ttft_ms": 60.15229190234095, "decode_ms": 148.79533310886472}, "total_wall_ms": 340.8499580109492, "ttft_ms": 60.15229190234095, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 484868096, "backend_memory": {"mlx_active_bytes": 6268821512, "mlx_peak_bytes": 6306619316, "mlx_cache_bytes": 3582855323, "process_rss_bytes": 484868096}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6306619316, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c3d5da842f16f311212d8a70f4b9b00acfc49bad8c8a6407cc8219c2e896e52d", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}185{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "7ea0033c46473df9d3ee7fc1", "attempt_id": "9836c20c488c", "run_group_id": "E1_pilot:B5|keep_frac=0.125|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:10.410761Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.125, "access": "aware", "realized": {"policy": "attention", "keep_frac": 0.125, "T": 2010, "keep": 251, "window_tokens": 46, "compress_ms": 184.51749999076128}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9252864, "serialized_payload_bytes": 9268749, "side_information_bytes": 15885, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 190.18683396279812, "doc_prefill_ms": 978.4983330173418, "compress_ms": 186.27745797857642, "serialize_ms": 3.599500050768256, "load_ms": 0.9678340284153819, "receiver_prefill_ms": 0.0, "query_prefill_ms": 28.286790940910578, "ttft_ms": 28.286790940910578, "decode_ms": 278.8742919219658}, "total_wall_ms": 500.47645904123783, "ttft_ms": 28.286790940910578, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 484900864, "backend_memory": {"mlx_active_bytes": 6268821512, "mlx_peak_bytes": 6311649224, "mlx_cache_bytes": 3582855323, "process_rss_bytes": 484900864}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6311649224, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6883757", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c3d5da842f16f311212d8a70f4b9b00acfc49bad8c8a6407cc8219c2e896e52d", "output_hash": "4f4163e571535bafb510d80d8bb642f2da7112be39e6271e34e285925abe4969"}186{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "fc2562a4f3375f98ee156d96", "attempt_id": "7f7229f935ed", "run_group_id": "E1_pilot:B5|keep_frac=0.25|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:11.089293Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "blind", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 19, "compress_ms": 136.959416908212}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521684, "side_information_bytes": 15956, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 153.9095420157537, "doc_prefill_ms": 978.4983330173418, "compress_ms": 143.25095899403095, "serialize_ms": 10.435374919325113, "load_ms": 1.522707985714078, "receiver_prefill_ms": 0.0, "query_prefill_ms": 34.654041985049844, "ttft_ms": 34.654041985049844, "decode_ms": 41.85833293013275}, "total_wall_ms": 79.97604098636657, "ttft_ms": 34.654041985049844, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 486572032, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6321594572, "mlx_cache_bytes": 3586312347, "process_rss_bytes": 486572032}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6321594572, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "24f23031bd1182e38854cebc49b2df6c094404387109ed4a2e330fca479f883a", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}187{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "17802194fa0119509b7e3b59", "attempt_id": "ec101ed582ce", "run_group_id": "E1_pilot:B5|keep_frac=0.25|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:11.195927Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "blind", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 19, "compress_ms": 136.959416908212}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521684, "side_information_bytes": 15956, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 153.9095420157537, "doc_prefill_ms": 978.4983330173418, "compress_ms": 143.25095899403095, "serialize_ms": 10.435374919325113, "load_ms": 1.6749589703977108, "receiver_prefill_ms": 0.0, "query_prefill_ms": 51.49516696110368, "ttft_ms": 51.49516696110368, "decode_ms": 36.127666011452675}, "total_wall_ms": 91.19587496388704, "ttft_ms": 51.49516696110368, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 486572032, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6327214308, "mlx_cache_bytes": 3586312347, "process_rss_bytes": 486572032}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327214308, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "24f23031bd1182e38854cebc49b2df6c094404387109ed4a2e330fca479f883a", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}188{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "47a7273716e82cdc5f65d69c", "attempt_id": "c2a787edb06f", "run_group_id": "E1_pilot:B5|keep_frac=0.25|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:11.312939Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "blind", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 19, "compress_ms": 136.959416908212}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521684, "side_information_bytes": 15956, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 153.9095420157537, "doc_prefill_ms": 978.4983330173418, "compress_ms": 143.25095899403095, "serialize_ms": 10.435374919325113, "load_ms": 3.8703340105712414, "receiver_prefill_ms": 0.0, "query_prefill_ms": 43.67375001311302, "ttft_ms": 43.67375001311302, "decode_ms": 24.828667053952813}, "total_wall_ms": 73.93125002272427, "ttft_ms": 43.67375001311302, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 486572032, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6327214308, "mlx_cache_bytes": 3586312347, "process_rss_bytes": 486572032}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6327214308, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "24f23031bd1182e38854cebc49b2df6c094404387109ed4a2e330fca479f883a", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}189{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ebfe35e2d6e8d7620ee332c7", "attempt_id": "dd13e6cfc62c", "run_group_id": "E1_pilot:B5|keep_frac=0.25|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:11.409493Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "blind", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 19, "compress_ms": 136.959416908212}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521684, "side_information_bytes": 15956, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 153.9095420157537, "doc_prefill_ms": 978.4983330173418, "compress_ms": 143.25095899403095, "serialize_ms": 10.435374919325113, "load_ms": 1.6111669829115272, "receiver_prefill_ms": 0.0, "query_prefill_ms": 47.31049994006753, "ttft_ms": 47.31049994006753, "decode_ms": 22.76508300565183}, "total_wall_ms": 73.38266691658646, "ttft_ms": 47.31049994006753, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 486572032, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6324396260, "mlx_cache_bytes": 3586312347, "process_rss_bytes": 486572032}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6324396260, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "24f23031bd1182e38854cebc49b2df6c094404387109ed4a2e330fca479f883a", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}190{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2e5514e6150a03d8b51a73bf", "attempt_id": "ad400896a517", "run_group_id": "E1_pilot:B5|keep_frac=0.25|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:11.506436Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "blind", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 19, "compress_ms": 136.959416908212}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521684, "side_information_bytes": 15956, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 153.9095420157537, "doc_prefill_ms": 978.4983330173418, "compress_ms": 143.25095899403095, "serialize_ms": 10.435374919325113, "load_ms": 6.6071669571101665, "receiver_prefill_ms": 0.0, "query_prefill_ms": 53.84733295068145, "ttft_ms": 53.84733295068145, "decode_ms": 326.44458406139165}, "total_wall_ms": 388.4931670036167, "ttft_ms": 53.84733295068145, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 486588416, "backend_memory": {"mlx_active_bytes": 6287548424, "mlx_peak_bytes": 6332375288, "mlx_cache_bytes": 3585067163, "process_rss_bytes": 486588416}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6332375288, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6620744", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "24f23031bd1182e38854cebc49b2df6c094404387109ed4a2e330fca479f883a", "output_hash": "e76b03e6d080dbb985ddc2cf22096423dd61750c5e72e827372501dee02ceb4e"}191{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3e142b5be6d7cf9c6be18dbc", "attempt_id": "522dbe145223", "run_group_id": "E1_pilot:B6|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:11.917359Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B6", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "pyramid", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 35, "compress_ms": 115.52699992898852}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521634, "side_information_bytes": 15906, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 124.4567499961704, "doc_prefill_ms": 978.4983330173418, "compress_ms": 117.3142910702154, "serialize_ms": 6.888333009555936, "load_ms": 1.62529107183218, "receiver_prefill_ms": 0.0, "query_prefill_ms": 800.1455829944462, "ttft_ms": 800.1455829944462, "decode_ms": 350.3892079461366}, "total_wall_ms": 1278.5456669516861, "ttft_ms": 800.1455829944462, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 8}, "provider_cached_usage": null, "process_memory": 542736384, "backend_memory": {"mlx_active_bytes": 6287466504, "mlx_peak_bytes": 6315139880, "mlx_cache_bytes": 3672254107, "process_rss_bytes": 542736384}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6315139880, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-141414", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f5fe19c286e3295a2b76fb4e8edfc381f8a46a74af827a4976a1f8203ec5727", "output_hash": "3accb8103c2ea6bde33116666e47bc4d4516c476bf679884da62e6e0a833dc04"}192{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2fb8d2d7868b9757b6906b45", "attempt_id": "5bdc202dcc63", "run_group_id": "E1_pilot:B6|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:13.219009Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B6", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "pyramid", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 122.03500000759959}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521634, "side_information_bytes": 15906, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 133.6194579489529, "doc_prefill_ms": 978.4983330173418, "compress_ms": 123.72708308976144, "serialize_ms": 9.604208054952323, "load_ms": 1.4203749597072601, "receiver_prefill_ms": 0.0, "query_prefill_ms": 1016.9714589137584, "ttft_ms": 1016.9714589137584, "decode_ms": 377.6179590495303}, "total_wall_ms": 1531.744375010021, "ttft_ms": 1016.9714589137584, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 566542336, "backend_memory": {"mlx_active_bytes": 6287745032, "mlx_peak_bytes": 6316676996, "mlx_cache_bytes": 3672223043, "process_rss_bytes": 566542336}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6316676996, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f5fe19c286e3295a2b76fb4e8edfc381f8a46a74af827a4976a1f8203ec5727", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}193{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "152b1c4a3d2614c7bea042f1", "attempt_id": "208d0d77bb52", "run_group_id": "E1_pilot:B6|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:14.775180Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B6", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "pyramid", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 100.8127499371767}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521633, "side_information_bytes": 15905, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 112.16529202647507, "doc_prefill_ms": 978.4983330173418, "compress_ms": 102.71037498023361, "serialize_ms": 9.15120798163116, "load_ms": 0.5802500527352095, "receiver_prefill_ms": 0.0, "query_prefill_ms": 948.3068339759484, "ttft_ms": 948.3068339759484, "decode_ms": 304.3392919935286}, "total_wall_ms": 1367.4807919887826, "ttft_ms": 948.3068339759484, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 581877760, "backend_memory": {"mlx_active_bytes": 6287679496, "mlx_peak_bytes": 6317130848, "mlx_cache_bytes": 3672288579, "process_rss_bytes": 581877760}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6317130848, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "1c6b2147d42e029e8dca07f4b28e03a278fb7e32227e176a30a7d1d298e4cef0", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}194{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "848c8584aba45ac3f3289a06", "attempt_id": "1c39334b216b", "run_group_id": "E1_pilot:B6|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:16.168373Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B6", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "pyramid", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 38, "compress_ms": 81.36487496085465}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521633, "side_information_bytes": 15905, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 91.60854096990079, "doc_prefill_ms": 978.4983330173418, "compress_ms": 83.10245897155255, "serialize_ms": 8.180457982234657, "load_ms": 1.1518340324983, "receiver_prefill_ms": 0.0, "query_prefill_ms": 967.3365419730544, "ttft_ms": 967.3365419730544, "decode_ms": 192.0231250114739}, "total_wall_ms": 1254.1444998933002, "ttft_ms": 967.3365419730544, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 596230144, "backend_memory": {"mlx_active_bytes": 6287417352, "mlx_peak_bytes": 6316319792, "mlx_cache_bytes": 3674647875, "process_rss_bytes": 596230144}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6316319792, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "1c6b2147d42e029e8dca07f4b28e03a278fb7e32227e176a30a7d1d298e4cef0", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}195{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "dc4d7ab44ba555ab3deea38f", "attempt_id": "635d72cfddd5", "run_group_id": "E1_pilot:B6|keep_frac=0.25|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:17.446601Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B6", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"keep_frac": 0.25, "access": "aware", "realized": {"policy": "pyramid", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 46, "compress_ms": 156.07712499331683}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 18505728, "serialized_payload_bytes": 18521634, "side_information_bytes": 15906, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 164.31133402511477, "doc_prefill_ms": 978.4983330173418, "compress_ms": 157.64266694895923, "serialize_ms": 6.35245896410197, "load_ms": 1.0650829644873738, "receiver_prefill_ms": 0.0, "query_prefill_ms": 1094.7622910607606, "ttft_ms": 1094.7622910607606, "decode_ms": 355.96962505951524}, "total_wall_ms": 1617.846457986161, "ttft_ms": 1094.7622910607606, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 613564416, "backend_memory": {"mlx_active_bytes": 6287908872, "mlx_peak_bytes": 6317799400, "mlx_cache_bytes": 3674395343, "process_rss_bytes": 613564416}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6317799400, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,882,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7f5fe19c286e3295a2b76fb4e8edfc381f8a46a74af827a4976a1f8203ec5727", "output_hash": "1886eda11e703e25a27facb91bdc6cb3774b1920e65fa8d88ed4639d56209253"}196{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8f2b5bc28013e60d4a4330cc", "attempt_id": "65e6da0d8219", "run_group_id": "E1_pilot:B9|bits=8|keep_frac=0.25|policy=attention|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:19.090662Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B9", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "bits": 8, "access": "aware", "policy": "attention", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 35, "compress_ms": 201.04399998672307, "bits": 8}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9831168, "serialized_payload_bytes": 9864690, "side_information_bytes": 33522, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 221.37945797294378, "doc_prefill_ms": 978.4983330173418, "compress_ms": 209.87866609357297, "serialize_ms": 10.782083962112665, "load_ms": 3.587624989449978, "receiver_prefill_ms": 0.0, "query_prefill_ms": 60.996292042545974, "ttft_ms": 60.996292042545974, "decode_ms": 313.94500005990267}, "total_wall_ms": 602.8465839335695, "ttft_ms": 60.996292042545974, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 35}, "output_usage": {"output_tokens": 6}, "provider_cached_usage": null, "process_memory": 610566144, "backend_memory": {"mlx_active_bytes": 6268717192, "mlx_peak_bytes": 6309297656, "mlx_cache_bytes": 3703190263, "process_rss_bytes": 610566144}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6309297656, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-1482", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7e9223c5763d06ad8b79c2db726ebbf118587b3dce8be8a16214375cf60174fc", "output_hash": "f5e9a6d78321ae31e0a3bdeeaf8c39b796c1fc12a0a8896955e8c411f428367c"}197{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "793887e91f2b0fafb391da85", "attempt_id": "0e08344cf658", "run_group_id": "E1_pilot:B9|bits=8|keep_frac=0.25|policy=attention|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:19.724698Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B9", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "bits": 8, "access": "aware", "policy": "attention", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 96.76933393348008, "bits": 8}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9831168, "serialized_payload_bytes": 9864689, "side_information_bytes": 33521, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 159.62045802734792, "doc_prefill_ms": 978.4983330173418, "compress_ms": 153.05966697633266, "serialize_ms": 6.034082965925336, "load_ms": 3.6637920420616865, "receiver_prefill_ms": 0.0, "query_prefill_ms": 54.22699998598546, "ttft_ms": 54.22699998598546, "decode_ms": 410.92024999670684}, "total_wall_ms": 630.353125045076, "ttft_ms": 54.22699998598546, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 611729408, "backend_memory": {"mlx_active_bytes": 6268720512, "mlx_peak_bytes": 6314674528, "mlx_cache_bytes": 3714773159, "process_rss_bytes": 611729408}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6314674528, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "eac92062bcbe3818105a60e3bbe6293aa5a070ab1c5f619945287aae7244bcfc", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}198{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f00636fb7bdc9c02383bd003", "attempt_id": "774438419334", "run_group_id": "E1_pilot:B9|bits=8|keep_frac=0.25|policy=attention|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:20.385890Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B9", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "bits": 8, "access": "aware", "policy": "attention", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 170.21425010170788, "bits": 8}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9831168, "serialized_payload_bytes": 9864690, "side_information_bytes": 33522, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 189.20416699256748, "doc_prefill_ms": 978.4983330173418, "compress_ms": 180.0569590413943, "serialize_ms": 8.405499975197017, "load_ms": 1.6090830322355032, "receiver_prefill_ms": 0.0, "query_prefill_ms": 67.88970797788352, "ttft_ms": 67.88970797788352, "decode_ms": 309.36858302447945}, "total_wall_ms": 570.77833393123, "ttft_ms": 67.88970797788352, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 41}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 611762176, "backend_memory": {"mlx_active_bytes": 6268646784, "mlx_peak_bytes": 6314601720, "mlx_cache_bytes": 3714846887, "process_rss_bytes": 611762176}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6314601720, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7e9223c5763d06ad8b79c2db726ebbf118587b3dce8be8a16214375cf60174fc", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}199{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "cd86b829645e0f61fb112b9f", "attempt_id": "2ad10bf5f188", "run_group_id": "E1_pilot:B9|bits=8|keep_frac=0.25|policy=attention|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:20.985445Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B9", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "bits": 8, "access": "aware", "policy": "attention", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 38, "compress_ms": 117.36891604959965, "bits": 8}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9831168, "serialized_payload_bytes": 9864690, "side_information_bytes": 33522, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 167.5742919323966, "doc_prefill_ms": 978.4983330173418, "compress_ms": 160.52779206074774, "serialize_ms": 6.497083115391433, "load_ms": 3.965583979152143, "receiver_prefill_ms": 0.0, "query_prefill_ms": 53.30499995034188, "ttft_ms": 53.30499995034188, "decode_ms": 221.48974996525794}, "total_wall_ms": 448.55570793151855, "ttft_ms": 53.30499995034188, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 38}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 611827712, "backend_memory": {"mlx_active_bytes": 6268641128, "mlx_peak_bytes": 6311321464, "mlx_cache_bytes": 3720898239, "process_rss_bytes": 611827712}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6311321464, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "7e9223c5763d06ad8b79c2db726ebbf118587b3dce8be8a16214375cf60174fc", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}200{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8567987fcf7d3ef563754f89", "attempt_id": "6df972514b29", "run_group_id": "E1_pilot:B9|bits=8|keep_frac=0.25|policy=attention|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:21.463912Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B9", "implementation_fidelity": "custom", "compression_config": {"keep_frac": 0.25, "bits": 8, "access": "aware", "policy": "attention", "realized": {"policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 46, "compress_ms": 91.65679104626179, "bits": 8}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 9831168, "serialized_payload_bytes": 9864689, "side_information_bytes": 33521, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 107.72583400830626, "doc_prefill_ms": 978.4983330173418, "compress_ms": 99.87862501293421, "serialize_ms": 7.305750041268766, "load_ms": 1.2198749464005232, "receiver_prefill_ms": 0.0, "query_prefill_ms": 52.16600000858307, "ttft_ms": 52.16600000858307, "decode_ms": 349.2728329729289}, "total_wall_ms": 512.5126250786707, "ttft_ms": 52.16600000858307, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 74096640, "query_tokens": 46}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 612171776, "backend_memory": {"mlx_active_bytes": 6268716160, "mlx_peak_bytes": 6319394620, "mlx_cache_bytes": 3730092375, "process_rss_bytes": 612171776}, "peak_prefill_memory": 7048282024, "peak_decode_memory": 6319394620, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6889172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "eac92062bcbe3818105a60e3bbe6293aa5a070ab1c5f619945287aae7244bcfc", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}201{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a284ef353d3769af31d73449", "attempt_id": "0b615e4bea1e", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:27.093179Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597378376}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 86, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.03316707443445921, "decode_payload_ms": 0.0, "receiver_prefill_ms": 646.6909170849249, "ttft_ms": 47.980832983739674, "decode_ms": 36.75787488464266}, "total_wall_ms": 746.2564589222893, "ttft_ms": 47.980832983739674, "input_usage": {"receiver_prompt_tokens": 86}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3889004544, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12684539768, "mlx_cache_bytes": 163771776, "process_rss_bytes": 3889004544}, "peak_prefill_memory": null, "peak_decode_memory": 12684539768, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b764cdc0eab7137467211272fa539f1260d1bf2e71bcf6ff3bdc960f5c16aa14"}202{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a7d32ca67ec7846b7a03ff38", "attempt_id": "d65cab2b1970", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:27.893656Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 91, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.03316707443445921, "decode_payload_ms": 0.0, "receiver_prefill_ms": 53.04708401672542, "ttft_ms": 21.096749929711223, "decode_ms": 77.90612510871142}, "total_wall_ms": 153.12387503217906, "ttft_ms": 21.096749929711223, "input_usage": {"receiver_prompt_tokens": 91}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 3888316416, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12688086476, "mlx_cache_bytes": 226719464, "process_rss_bytes": 3888316416}, "peak_prefill_memory": null, "peak_decode_memory": 12688086476, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Unknown.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "9cf9f50925b2a7803c73826a305235d4215dc428721c731072a9f2c0f5d6ae5a"}203{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c1fce208a1c64cf5e6afee35", "attempt_id": "93cc7bc4dbc0", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:28.080446Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 91, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.03316707443445921, "decode_payload_ms": 0.0, "receiver_prefill_ms": 51.64225003682077, "ttft_ms": 38.24829205404967, "decode_ms": 76.70120801776648}, "total_wall_ms": 167.41258301772177, "ttft_ms": 38.24829205404967, "input_usage": {"receiver_prompt_tokens": 91}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 3888332800, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12688086476, "mlx_cache_bytes": 226719464, "process_rss_bytes": 3888332800}, "peak_prefill_memory": null, "peak_decode_memory": 12688086476, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Unknown.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "9cf9f50925b2a7803c73826a305235d4215dc428721c731072a9f2c0f5d6ae5a"}204{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9ec1c73d4a4e9d72cde7212b", "attempt_id": "6fd4645d9a5f", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:28.277113Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 88, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.03316707443445921, "decode_payload_ms": 0.0, "receiver_prefill_ms": 66.67800003197044, "ttft_ms": 34.88487494178116, "decode_ms": 84.5717079937458}, "total_wall_ms": 187.05820792820305, "ttft_ms": 34.88487494178116, "input_usage": {"receiver_prompt_tokens": 88}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 3888349184, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12691871180, "mlx_cache_bytes": 253163240, "process_rss_bytes": 3888349184}, "peak_prefill_memory": null, "peak_decode_memory": 12691871180, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Unknown.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "9cf9f50925b2a7803c73826a305235d4215dc428721c731072a9f2c0f5d6ae5a"}205{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f32cc40904d767767c3ce27d", "attempt_id": "bc5e45b1ac68", "run_group_id": "E1_pilot:A0", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:28.492103Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A0", "implementation_fidelity": "custom", "compression_config": {"realized": {"note": "receiver gets no source context"}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 0, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 97, "message_tokens": 0, "message_utf8_bytes": 0, "cache_tensor_bytes": null, "serialized_payload_bytes": 0, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.03316707443445921, "decode_payload_ms": 0.0, "receiver_prefill_ms": 45.02454097382724, "ttft_ms": 29.383999994024634, "decode_ms": 38.97570900153369}, "total_wall_ms": 114.28587499540299, "ttft_ms": 29.383999994024634, "input_usage": {"receiver_prompt_tokens": 97}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3888398336, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12691756516, "mlx_cache_bytes": 315684968, "process_rss_bytes": 3888398336}, "peak_prefill_memory": null, "peak_decode_memory": 12691756516, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "output_hash": "b764cdc0eab7137467211272fa539f1260d1bf2e71bcf6ff3bdc960f5c16aa14"}206{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3396758962deff05a4c73dcd", "attempt_id": "4cbfbfb37fc2", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:28.634823Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1862, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01770805101841688, "decode_payload_ms": 0.0, "receiver_prefill_ms": 499.6396249625832, "ttft_ms": 43.695083004422486, "decode_ms": 249.3790000444278}, "total_wall_ms": 796.4147080201656, "ttft_ms": 43.695083004422486, "input_usage": {"receiver_prompt_tokens": 1862}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 3885826048, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13292777800, "mlx_cache_bytes": 1022268796, "process_rss_bytes": 3885826048}, "peak_prefill_memory": null, "peak_decode_memory": 13292777800, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}207{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "fb59f1bf107d1bf128a66a27", "attempt_id": "b34d5c0a288f", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:29.461431Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1867, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01770805101841688, "decode_payload_ms": 0.0, "receiver_prefill_ms": 579.509292030707, "ttft_ms": 55.44162495061755, "decode_ms": 322.7411250118166}, "total_wall_ms": 961.2702080048621, "ttft_ms": 55.44162495061755, "input_usage": {"receiver_prompt_tokens": 1867}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 3885842432, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13294137692, "mlx_cache_bytes": 1492529828, "process_rss_bytes": 3885842432}, "peak_prefill_memory": null, "peak_decode_memory": 13294137692, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "5280344bb130ba2ce68331d4c5deadfa47c813ecd5c007f5616e1ff44aa0daea"}208{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3a92313afb8a49078d325ead", "attempt_id": "5234e5198298", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:30.450859Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1867, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01770805101841688, "decode_payload_ms": 0.0, "receiver_prefill_ms": 476.67066601570696, "ttft_ms": 62.968041049316525, "decode_ms": 180.50954199861735}, "total_wall_ms": 723.2565829763189, "ttft_ms": 62.968041049316525, "input_usage": {"receiver_prompt_tokens": 1867}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3885842432, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13294137692, "mlx_cache_bytes": 1492529828, "process_rss_bytes": 3885842432}, "peak_prefill_memory": null, "peak_decode_memory": 13294137692, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}209{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "b3368f2891c0a96ee96f3a80", "attempt_id": "f1fa5ab919ef", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:31.199522Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1864, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01770805101841688, "decode_payload_ms": 0.0, "receiver_prefill_ms": 451.5241249464452, "ttft_ms": 42.268167017027736, "decode_ms": 200.49616694450378}, "total_wall_ms": 697.5888749584556, "ttft_ms": 42.268167017027736, "input_usage": {"receiver_prompt_tokens": 1864}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3885006848, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13331100008, "mlx_cache_bytes": 1732932260, "process_rss_bytes": 3885006848}, "peak_prefill_memory": null, "peak_decode_memory": 13331100008, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}210{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3c0afc9de90055c13bb3f607", "attempt_id": "ecb510fdc245", "run_group_id": "E1_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:31.923455Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1873, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 7280, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01770805101841688, "decode_payload_ms": 0.0, "receiver_prefill_ms": 429.7551669878885, "ttft_ms": 59.38220804091543, "decode_ms": 219.0751669695601}, "total_wall_ms": 712.5919159734622, "ttft_ms": 59.38220804091543, "input_usage": {"receiver_prompt_tokens": 1873}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3882303488, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13295448436, "mlx_cache_bytes": 2169999332, "process_rss_bytes": 3882303488}, "peak_prefill_memory": null, "peak_decode_memory": 13295448436, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}211{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f84c1e5c2124916cc6dbb84e", "attempt_id": "fb0b068f3044", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:32.663412Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 232, "message_tokens": 131, "message_utf8_bytes": 703, "cache_tensor_bytes": null, "serialized_payload_bytes": 703, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4802919942885637, "decode_payload_ms": 0.0, "receiver_prefill_ms": 100.58099997695535, "ttft_ms": 26.880832971073687, "decode_ms": 25.019083986990154}, "total_wall_ms": 154.02700006961823, "ttft_ms": 26.880832971073687, "input_usage": {"receiver_prompt_tokens": 232}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3882270720, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12759980020, "mlx_cache_bytes": 2288112512, "process_rss_bytes": 3882270720}, "peak_prefill_memory": null, "peak_decode_memory": 12759980020, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "acba6fb02b6e7258fce5128bcba1f258105bb31a37756a67b631e93f89e9cb50", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}212{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ee94c3b9667dec1f99dcc6e1", "attempt_id": "dafd70d2167b", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:32.851298Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 237, "message_tokens": 131, "message_utf8_bytes": 703, "cache_tensor_bytes": null, "serialized_payload_bytes": 703, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4802919942885637, "decode_payload_ms": 0.0, "receiver_prefill_ms": 85.92129102908075, "ttft_ms": 46.80308303795755, "decode_ms": 36.294416058808565}, "total_wall_ms": 170.1134160393849, "ttft_ms": 46.80308303795755, "input_usage": {"receiver_prompt_tokens": 237}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3878420480, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12767565832, "mlx_cache_bytes": 2414679856, "process_rss_bytes": 3878420480}, "peak_prefill_memory": null, "peak_decode_memory": 12767565832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "acba6fb02b6e7258fce5128bcba1f258105bb31a37756a67b631e93f89e9cb50", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}213{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "594463cf469011efdf778ecc", "attempt_id": "095a9a8290c3", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:33.044656Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 237, "message_tokens": 131, "message_utf8_bytes": 703, "cache_tensor_bytes": null, "serialized_payload_bytes": 703, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4802919942885637, "decode_payload_ms": 0.0, "receiver_prefill_ms": 84.47579201310873, "ttft_ms": 38.24570798315108, "decode_ms": 37.18833299353719}, "total_wall_ms": 161.05399990919977, "ttft_ms": 38.24570798315108, "input_usage": {"receiver_prompt_tokens": 237}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3878256640, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12767565832, "mlx_cache_bytes": 2414679856, "process_rss_bytes": 3878256640}, "peak_prefill_memory": null, "peak_decode_memory": 12767565832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "acba6fb02b6e7258fce5128bcba1f258105bb31a37756a67b631e93f89e9cb50", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}214{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "479d02184807ec5f02abd94a", "attempt_id": "b3ad3df4d752", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:33.230295Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 234, "message_tokens": 131, "message_utf8_bytes": 703, "cache_tensor_bytes": null, "serialized_payload_bytes": 703, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4802919942885637, "decode_payload_ms": 0.0, "receiver_prefill_ms": 94.60470802150667, "ttft_ms": 37.55633404944092, "decode_ms": 36.75004199612886}, "total_wall_ms": 170.09266698732972, "ttft_ms": 37.55633404944092, "input_usage": {"receiver_prompt_tokens": 234}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3878256640, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12770940936, "mlx_cache_bytes": 2440632112, "process_rss_bytes": 3878256640}, "peak_prefill_memory": null, "peak_decode_memory": 12770940936, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "acba6fb02b6e7258fce5128bcba1f258105bb31a37756a67b631e93f89e9cb50", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}215{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "edb7327bfa87914681f8ec58", "attempt_id": "bfc302a6223a", "run_group_id": "E1_pilot:A2|budget_tokens=128", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:33.427201Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "realized": {"budget_tokens": 128, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 243, "message_tokens": 131, "message_utf8_bytes": 703, "cache_tensor_bytes": null, "serialized_payload_bytes": 703, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4802919942885637, "decode_payload_ms": 0.0, "receiver_prefill_ms": 79.13520804140717, "ttft_ms": 31.041332986205816, "decode_ms": 31.99995798058808}, "total_wall_ms": 143.47324997652322, "ttft_ms": 31.041332986205816, "input_usage": {"receiver_prompt_tokens": 243}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3861282816, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12766648352, "mlx_cache_bytes": 2576112376, "process_rss_bytes": 3861282816}, "peak_prefill_memory": null, "peak_decode_memory": 12766648352, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "acba6fb02b6e7258fce5128bcba1f258105bb31a37756a67b631e93f89e9cb50", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}216{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ba0f7700e5edd41000a20d02", "attempt_id": "5ee8bde226ea", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:33.596034Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 617, "message_tokens": 516, "message_utf8_bytes": 1940, "cache_tensor_bytes": null, "serialized_payload_bytes": 1940, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.340000075288117, "decode_payload_ms": 0.0, "receiver_prefill_ms": 173.50241693202406, "ttft_ms": 40.39966594427824, "decode_ms": 49.36187504790723}, "total_wall_ms": 265.28537506237626, "ttft_ms": 40.39966594427824, "input_usage": {"receiver_prompt_tokens": 617}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3860496384, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13039345144, "mlx_cache_bytes": 2963874968, "process_rss_bytes": 3860496384}, "peak_prefill_memory": null, "peak_decode_memory": 13039345144, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "83c1970dac87ef7396a52d14a2764a080806f090cf3b37207beb09b20af11f42", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}217{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "aa329def04e2af8950e813be", "attempt_id": "030e054b0b4c", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:33.892164Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 622, "message_tokens": 516, "message_utf8_bytes": 1940, "cache_tensor_bytes": null, "serialized_payload_bytes": 1940, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.340000075288117, "decode_payload_ms": 0.0, "receiver_prefill_ms": 249.12733398377895, "ttft_ms": 56.85391603037715, "decode_ms": 325.82404208369553}, "total_wall_ms": 633.8349159341305, "ttft_ms": 56.85391603037715, "input_usage": {"receiver_prompt_tokens": 622}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 3856957440, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13050306060, "mlx_cache_bytes": 3277008460, "process_rss_bytes": 3856957440}, "peak_prefill_memory": null, "peak_decode_memory": 13050306060, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "83c1970dac87ef7396a52d14a2764a080806f090cf3b37207beb09b20af11f42", "output_hash": "5280344bb130ba2ce68331d4c5deadfa47c813ecd5c007f5616e1ff44aa0daea"}218{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "10191d7636d6f2feeebd57e2", "attempt_id": "9e2e33cb3293", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:34.555307Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 622, "message_tokens": 516, "message_utf8_bytes": 1940, "cache_tensor_bytes": null, "serialized_payload_bytes": 1940, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.340000075288117, "decode_payload_ms": 0.0, "receiver_prefill_ms": 187.84987495746464, "ttft_ms": 49.86991605255753, "decode_ms": 34.08675000537187}, "total_wall_ms": 273.3824170427397, "ttft_ms": 49.86991605255753, "input_usage": {"receiver_prompt_tokens": 622}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3857006592, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13054139916, "mlx_cache_bytes": 3277008460, "process_rss_bytes": 3857006592}, "peak_prefill_memory": null, "peak_decode_memory": 13054139916, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "83c1970dac87ef7396a52d14a2764a080806f090cf3b37207beb09b20af11f42", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}219{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "cb128479d40644373de8a88a", "attempt_id": "f8c3d744a3f8", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:34.851168Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 619, "message_tokens": 516, "message_utf8_bytes": 1940, "cache_tensor_bytes": null, "serialized_payload_bytes": 1940, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.340000075288117, "decode_payload_ms": 0.0, "receiver_prefill_ms": 193.6789999017492, "ttft_ms": 36.55762493144721, "decode_ms": 48.74641599599272}, "total_wall_ms": 280.6981250178069, "ttft_ms": 36.55762493144721, "input_usage": {"receiver_prompt_tokens": 619}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3856744448, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13071900172, "mlx_cache_bytes": 3516280396, "process_rss_bytes": 3856744448}, "peak_prefill_memory": null, "peak_decode_memory": 13071900172, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "83c1970dac87ef7396a52d14a2764a080806f090cf3b37207beb09b20af11f42", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}220{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6607c938881b576685ecb266", "attempt_id": "c91bffda7e07", "run_group_id": "E1_pilot:A2|budget_tokens=512", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:35.156824Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A2", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "realized": {"budget_tokens": 512, "source_tokens": 1766}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 628, "message_tokens": 516, "message_utf8_bytes": 1940, "cache_tensor_bytes": null, "serialized_payload_bytes": 1940, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.340000075288117, "decode_payload_ms": 0.0, "receiver_prefill_ms": 223.83433301001787, "ttft_ms": 45.188624993897974, "decode_ms": 187.42116703651845}, "total_wall_ms": 457.8971250448376, "ttft_ms": 45.188624993897974, "input_usage": {"receiver_prompt_tokens": 628}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3780182016, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13046619684, "mlx_cache_bytes": 3844454424, "process_rss_bytes": 3780182016}, "peak_prefill_memory": null, "peak_decode_memory": 13046619684, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "83c1970dac87ef7396a52d14a2764a080806f090cf3b37207beb09b20af11f42", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}221{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "caa7b43d8b3eeab81cb80c2f", "attempt_id": "0dee985285ea", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:35.646836Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 225, "message_tokens": 128, "message_utf8_bytes": 396, "cache_tensor_bytes": null, "serialized_payload_bytes": 396, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4865410523489118, "decode_payload_ms": 0.0, "receiver_prefill_ms": 86.74400008749217, "ttft_ms": 27.555915992707014, "decode_ms": 41.38258297462016}, "total_wall_ms": 156.72666602768004, "ttft_ms": 27.555915992707014, "input_usage": {"receiver_prompt_tokens": 225}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3777789952, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12756047860, "mlx_cache_bytes": 3950426136, "process_rss_bytes": 3777789952}, "peak_prefill_memory": null, "peak_decode_memory": 12756047860, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f81418bc710665c0ce4e278485f28f6b252e7ffd1fac97fe3e5d9356dfee344d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}222{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "06a48ab2e01c81df4c6e29ea", "attempt_id": "4078075de022", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:35.830422Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 230, "message_tokens": 128, "message_utf8_bytes": 396, "cache_tensor_bytes": null, "serialized_payload_bytes": 396, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4865410523489118, "decode_payload_ms": 0.0, "receiver_prefill_ms": 74.36545891687274, "ttft_ms": 35.35516606643796, "decode_ms": 44.93241698946804}, "total_wall_ms": 155.7475000154227, "ttft_ms": 35.35516606643796, "input_usage": {"receiver_prompt_tokens": 230}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3777871872, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12770858996, "mlx_cache_bytes": 4009539608, "process_rss_bytes": 3777871872}, "peak_prefill_memory": null, "peak_decode_memory": 12770858996, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f81418bc710665c0ce4e278485f28f6b252e7ffd1fac97fe3e5d9356dfee344d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}223{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "25764a7d9afc2342a655efc1", "attempt_id": "1cf967571fa9", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:36.010841Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 230, "message_tokens": 128, "message_utf8_bytes": 396, "cache_tensor_bytes": null, "serialized_payload_bytes": 396, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4865410523489118, "decode_payload_ms": 0.0, "receiver_prefill_ms": 105.6317089824006, "ttft_ms": 46.735708019696176, "decode_ms": 46.452583046630025}, "total_wall_ms": 199.71116702072322, "ttft_ms": 46.735708019696176, "input_usage": {"receiver_prompt_tokens": 230}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3777904640, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12773709812, "mlx_cache_bytes": 4009539608, "process_rss_bytes": 3777904640}, "peak_prefill_memory": null, "peak_decode_memory": 12773709812, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f81418bc710665c0ce4e278485f28f6b252e7ffd1fac97fe3e5d9356dfee344d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}224{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "982dafb312e24a42a7792a60", "attempt_id": "63b2ddb5707e", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:36.233991Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 227, "message_tokens": 128, "message_utf8_bytes": 396, "cache_tensor_bytes": null, "serialized_payload_bytes": 396, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4865410523489118, "decode_payload_ms": 0.0, "receiver_prefill_ms": 100.63695895951241, "ttft_ms": 43.78033406101167, "decode_ms": 126.4765829546377}, "total_wall_ms": 272.0002500573173, "ttft_ms": 43.78033406101167, "input_usage": {"receiver_prompt_tokens": 227}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3777937408, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12760012788, "mlx_cache_bytes": 4106827800, "process_rss_bytes": 3777937408}, "peak_prefill_memory": null, "peak_decode_memory": 12760012788, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f81418bc710665c0ce4e278485f28f6b252e7ffd1fac97fe3e5d9356dfee344d", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}225{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "51cc0f3cf36f50b79791095d", "attempt_id": "559d9339fb47", "run_group_id": "E1_pilot:A3|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:36.530163Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "realized_tokens": 128, "n_sentences": 4}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 236, "message_tokens": 128, "message_utf8_bytes": 396, "cache_tensor_bytes": null, "serialized_payload_bytes": 396, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.4865410523489118, "decode_payload_ms": 0.0, "receiver_prefill_ms": 94.85150000546128, "ttft_ms": 38.70000003371388, "decode_ms": 41.803917032666504}, "total_wall_ms": 176.40208394732326, "ttft_ms": 38.70000003371388, "input_usage": {"receiver_prompt_tokens": 236}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3777953792, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12767565832, "mlx_cache_bytes": 4106827800, "process_rss_bytes": 3777953792}, "peak_prefill_memory": null, "peak_decode_memory": 12767565832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "f81418bc710665c0ce4e278485f28f6b252e7ffd1fac97fe3e5d9356dfee344d", "output_hash": "b764cdc0eab7137467211272fa539f1260d1bf2e71bcf6ff3bdc960f5c16aa14"}226{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "345f18ec5689466c3d8bad53", "attempt_id": "5c25b803c0ba", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:36.734103Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 509, "n_sentences": 18}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 606, "message_tokens": 509, "message_utf8_bytes": 1536, "cache_tensor_bytes": null, "serialized_payload_bytes": 1536, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.1802500123158097, "decode_payload_ms": 0.0, "receiver_prefill_ms": 176.01558403111994, "ttft_ms": 40.34449998289347, "decode_ms": 162.25595795549452}, "total_wall_ms": 384.94466699194163, "ttft_ms": 40.34449998289347, "input_usage": {"receiver_prompt_tokens": 606}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3778215936, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13033561592, "mlx_cache_bytes": 4445272088, "process_rss_bytes": 3778215936}, "peak_prefill_memory": null, "peak_decode_memory": 13033561592, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Haruki Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0ca1d3fed207a8962d88341b7fb1b07da00df9d8e507d786a2e99c06b18d0a62", "output_hash": "55a01495263476a6da84eabe2c22899f74b3909040b9d9d4fe8a8b6431ff9e18"}227{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "652a3ae76e38721a35e2b8f6", "attempt_id": "c767a49c5937", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:37.143386Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 509, "n_sentences": 18}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 611, "message_tokens": 509, "message_utf8_bytes": 1536, "cache_tensor_bytes": null, "serialized_payload_bytes": 1536, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.1802500123158097, "decode_payload_ms": 0.0, "receiver_prefill_ms": 183.93062497489154, "ttft_ms": 38.269583019427955, "decode_ms": 48.33074996713549}, "total_wall_ms": 272.01195899397135, "ttft_ms": 38.269583019427955, "input_usage": {"receiver_prompt_tokens": 611}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3778232320, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13036264952, "mlx_cache_bytes": 4741478424, "process_rss_bytes": 3778232320}, "peak_prefill_memory": null, "peak_decode_memory": 13036264952, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0ca1d3fed207a8962d88341b7fb1b07da00df9d8e507d786a2e99c06b18d0a62", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}228{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "cfad773caf5f462cf4ce6dad", "attempt_id": "efc4d182688c", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:37.441327Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 509, "n_sentences": 18}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 611, "message_tokens": 509, "message_utf8_bytes": 1536, "cache_tensor_bytes": null, "serialized_payload_bytes": 1536, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.1802500123158097, "decode_payload_ms": 0.0, "receiver_prefill_ms": 185.83787500392646, "ttft_ms": 51.831708988174796, "decode_ms": 25.50999994855374}, "total_wall_ms": 264.68404196202755, "ttft_ms": 51.831708988174796, "input_usage": {"receiver_prompt_tokens": 611}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3778265088, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13036264952, "mlx_cache_bytes": 4741478424, "process_rss_bytes": 3778265088}, "peak_prefill_memory": null, "peak_decode_memory": 13036264952, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0ca1d3fed207a8962d88341b7fb1b07da00df9d8e507d786a2e99c06b18d0a62", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}229{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0ee10688dcd2714df2d4ef36", "attempt_id": "4b8ddd641f6a", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:37.730986Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 509, "n_sentences": 18}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 608, "message_tokens": 509, "message_utf8_bytes": 1536, "cache_tensor_bytes": null, "serialized_payload_bytes": 1536, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.1802500123158097, "decode_payload_ms": 0.0, "receiver_prefill_ms": 200.8648751070723, "ttft_ms": 30.372041976079345, "decode_ms": 157.36887499224395}, "total_wall_ms": 390.51133301109076, "ttft_ms": 30.372041976079345, "input_usage": {"receiver_prompt_tokens": 608}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3778281472, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13065575928, "mlx_cache_bytes": 4898175000, "process_rss_bytes": 3778281472}, "peak_prefill_memory": null, "peak_decode_memory": 13065575928, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0ca1d3fed207a8962d88341b7fb1b07da00df9d8e507d786a2e99c06b18d0a62", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}230{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "73d8fd31983f5859f556bee6", "attempt_id": "99c1622af39b", "run_group_id": "E1_pilot:A3|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:38.147011Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "realized_tokens": 509, "n_sentences": 18}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 617, "message_tokens": 509, "message_utf8_bytes": 1536, "cache_tensor_bytes": null, "serialized_payload_bytes": 1536, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2.1802500123158097, "decode_payload_ms": 0.0, "receiver_prefill_ms": 184.36279194429517, "ttft_ms": 42.966334032826126, "decode_ms": 208.3066669292748}, "total_wall_ms": 437.20804201439023, "ttft_ms": 42.966334032826126, "input_usage": {"receiver_prompt_tokens": 617}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3778297856, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13039345144, "mlx_cache_bytes": 4898175000, "process_rss_bytes": 3778297856}, "peak_prefill_memory": null, "peak_decode_memory": 13039345144, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0ca1d3fed207a8962d88341b7fb1b07da00df9d8e507d786a2e99c06b18d0a62", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}231{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "127e697cab82122cfd1999cf", "attempt_id": "c97f4425da9d", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:38.614839Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 511, "n_sentences": 28}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 608, "message_tokens": 511, "message_utf8_bytes": 1555, "cache_tensor_bytes": null, "serialized_payload_bytes": 1555, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.9690000442788005, "decode_payload_ms": 0.0, "receiver_prefill_ms": 146.1639580084011, "ttft_ms": 44.18754205107689, "decode_ms": 322.3565830849111}, "total_wall_ms": 515.3643330559134, "ttft_ms": 44.18754205107689, "input_usage": {"receiver_prompt_tokens": 608}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 3778625536, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13073079800, "mlx_cache_bytes": 4898175000, "process_rss_bytes": 3778625536}, "peak_prefill_memory": null, "peak_decode_memory": 13073079800, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0ff405ac03e5a10e14d572266d362d361b4219fa184f9be3bb806057879fd9ac", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}232{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "e7f66bd94321b2eec32831a9", "attempt_id": "5703b6664b58", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:39.154684Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 508, "n_sentences": 29}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 610, "message_tokens": 508, "message_utf8_bytes": 1715, "cache_tensor_bytes": null, "serialized_payload_bytes": 1715, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.1981669813394547, "decode_payload_ms": 0.0, "receiver_prefill_ms": 165.07095901761204, "ttft_ms": 41.04537493549287, "decode_ms": 297.8782909922302}, "total_wall_ms": 506.7767919972539, "ttft_ms": 41.04537493549287, "input_usage": {"receiver_prompt_tokens": 610}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 3778592768, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13036264952, "mlx_cache_bytes": 4898175000, "process_rss_bytes": 3778592768}, "peak_prefill_memory": null, "peak_decode_memory": 13036264952, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2a8e20793db30181404c83877c2fa0ce0e9f4ce155842796b2e9a4db57e4d166", "output_hash": "5280344bb130ba2ce68331d4c5deadfa47c813ecd5c007f5616e1ff44aa0daea"}233{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "997f1353002369abfe7f7d4c", "attempt_id": "a2488decd017", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:39.686960Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 512, "n_sentences": 29}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 614, "message_tokens": 512, "message_utf8_bytes": 1561, "cache_tensor_bytes": null, "serialized_payload_bytes": 1561, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.2128340313211083, "decode_payload_ms": 0.0, "receiver_prefill_ms": 186.9793749647215, "ttft_ms": 42.87391703110188, "decode_ms": 155.52100003696978}, "total_wall_ms": 388.1841250695288, "ttft_ms": 42.87391703110188, "input_usage": {"receiver_prompt_tokens": 614}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3778707456, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13068918264, "mlx_cache_bytes": 5056378904, "process_rss_bytes": 3778707456}, "peak_prefill_memory": null, "peak_decode_memory": 13068918264, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "a53881c8306c0d00c837f52a61f488dcff8e4df7502a72cc68c383b3156842b8", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}234{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6ef8d11d45aadd5fa261753b", "attempt_id": "469f86e51af5", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:40.105304Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 511, "n_sentences": 22}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 606, "message_tokens": 507, "message_utf8_bytes": 1893, "cache_tensor_bytes": null, "serialized_payload_bytes": 1893, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1.043542055413127, "decode_payload_ms": 0.0, "receiver_prefill_ms": 151.80062502622604, "ttft_ms": 44.20812497846782, "decode_ms": 223.0843750294298}, "total_wall_ms": 421.5364999836311, "ttft_ms": 44.20812497846782, "input_usage": {"receiver_prompt_tokens": 606}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3778756608, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13033561592, "mlx_cache_bytes": 5056378904, "process_rss_bytes": 3778756608}, "peak_prefill_memory": null, "peak_decode_memory": 13033561592, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "22d6a96c1f0de138d69e1c361d6afdd8664290d33341c244479d5699bb0376d5", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}235{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "db8e21dc878e707ee4f94344", "attempt_id": "c45844d9c1ef", "run_group_id": "E1_pilot:A3|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:40.549920Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A3", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "realized_tokens": 512, "n_sentences": 19}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 620, "message_tokens": 512, "message_utf8_bytes": 1482, "cache_tensor_bytes": null, "serialized_payload_bytes": 1482, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.9942909236997366, "decode_payload_ms": 0.0, "receiver_prefill_ms": 182.39616591017693, "ttft_ms": 42.17254102695733, "decode_ms": 160.28391709551215}, "total_wall_ms": 387.1555000077933, "ttft_ms": 42.17254102695733, "input_usage": {"receiver_prompt_tokens": 620}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3778772992, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13053779468, "mlx_cache_bytes": 5178079256, "process_rss_bytes": 3778772992}, "peak_prefill_memory": null, "peak_decode_memory": 13053779468, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c0695a138fc1f4f65051dbbd242a17ff0f5189f02881adece262d80283d3d1b0", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}236{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8d7b1495a8227eccbf31b2ce", "attempt_id": "053515853bc8", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:46.602802Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 228, "message_tokens": 128, "message_utf8_bytes": 344, "cache_tensor_bytes": null, "serialized_payload_bytes": 344, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5641.347291995771, "decode_payload_ms": 0.0, "receiver_prefill_ms": 100.08387500420213, "ttft_ms": 41.48470796644688, "decode_ms": 165.93720810487866}, "total_wall_ms": 309.24616602715105, "ttft_ms": 41.48470796644688, "input_usage": {"receiver_prompt_tokens": 228, "sender_input_tokens": 1900, "sender_output_tokens": 128, "sender_prefill_ms": 571.6930000344291, "sender_ttft_ms": 49.959042109549046, "sender_decode_ms": 5017.545499955304}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3773546496, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12772956148, "mlx_cache_bytes": 5679273412, "process_rss_bytes": 3773546496}, "peak_prefill_memory": null, "peak_decode_memory": 12772956148, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Haruki Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e9b17f1eafffe7360614cc9ddf30c320799b427f7ef537b91abb8afc168ebe6", "output_hash": "55a01495263476a6da84eabe2c22899f74b3909040b9d9d4fe8a8b6431ff9e18"}237{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "20173f556a1391560d44d210", "attempt_id": "9067cec23a6d", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:46.945983Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 233, "message_tokens": 128, "message_utf8_bytes": 344, "cache_tensor_bytes": null, "serialized_payload_bytes": 344, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5641.347291995771, "decode_payload_ms": 0.0, "receiver_prefill_ms": 117.037917021662, "ttft_ms": 35.58658401016146, "decode_ms": 21.81591698899865}, "total_wall_ms": 175.41499994695187, "ttft_ms": 35.58658401016146, "input_usage": {"receiver_prompt_tokens": 233, "sender_input_tokens": 1900, "sender_output_tokens": 128, "sender_prefill_ms": 571.6930000344291, "sender_ttft_ms": 49.959042109549046, "sender_decode_ms": 5017.545499955304}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3773562880, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12760160264, "mlx_cache_bytes": 5679273412, "process_rss_bytes": 3773562880}, "peak_prefill_memory": null, "peak_decode_memory": 12760160264, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e9b17f1eafffe7360614cc9ddf30c320799b427f7ef537b91abb8afc168ebe6", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}238{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a7a1040c9c8fad5c3d466497", "attempt_id": "013d0136d6d1", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:47.146957Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 233, "message_tokens": 128, "message_utf8_bytes": 344, "cache_tensor_bytes": null, "serialized_payload_bytes": 344, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5641.347291995771, "decode_payload_ms": 0.0, "receiver_prefill_ms": 87.82808296382427, "ttft_ms": 40.84858298301697, "decode_ms": 46.31200002040714}, "total_wall_ms": 175.95975007861853, "ttft_ms": 40.84858298301697, "input_usage": {"receiver_prompt_tokens": 233, "sender_input_tokens": 1900, "sender_output_tokens": 128, "sender_prefill_ms": 571.6930000344291, "sender_ttft_ms": 49.959042109549046, "sender_decode_ms": 5017.545499955304}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3773595648, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12760160264, "mlx_cache_bytes": 5679273412, "process_rss_bytes": 3773595648}, "peak_prefill_memory": null, "peak_decode_memory": 12760160264, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e9b17f1eafffe7360614cc9ddf30c320799b427f7ef537b91abb8afc168ebe6", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}239{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "9103f469dfe0791cfa5628c9", "attempt_id": "af36ecb8a30b", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:47.346798Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 230, "message_tokens": 128, "message_utf8_bytes": 344, "cache_tensor_bytes": null, "serialized_payload_bytes": 344, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5641.347291995771, "decode_payload_ms": 0.0, "receiver_prefill_ms": 96.61016694735736, "ttft_ms": 21.791375009343028, "decode_ms": 38.34487497806549}, "total_wall_ms": 157.78904093895108, "ttft_ms": 21.791375009343028, "input_usage": {"receiver_prompt_tokens": 230, "sender_input_tokens": 1900, "sender_output_tokens": 128, "sender_prefill_ms": 571.6930000344291, "sender_ttft_ms": 49.959042109549046, "sender_decode_ms": 5017.545499955304}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3773644800, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12773709812, "mlx_cache_bytes": 5679273412, "process_rss_bytes": 3773644800}, "peak_prefill_memory": null, "peak_decode_memory": 12773709812, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e9b17f1eafffe7360614cc9ddf30c320799b427f7ef537b91abb8afc168ebe6", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}240{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a51c2dd7865a131d571756dd", "attempt_id": "905a906043f9", "run_group_id": "E1_pilot:A4|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:00:47.527386Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 239, "message_tokens": 128, "message_utf8_bytes": 344, "cache_tensor_bytes": null, "serialized_payload_bytes": 344, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5641.347291995771, "decode_payload_ms": 0.0, "receiver_prefill_ms": 108.96616708487272, "ttft_ms": 60.18854200374335, "decode_ms": 168.29575004521757}, "total_wall_ms": 338.52091699372977, "ttft_ms": 60.18854200374335, "input_usage": {"receiver_prompt_tokens": 239, "sender_input_tokens": 1900, "sender_output_tokens": 128, "sender_prefill_ms": 571.6930000344291, "sender_ttft_ms": 49.959042109549046, "sender_decode_ms": 5017.545499955304}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3773710336, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12778133536, "mlx_cache_bytes": 5714367940, "process_rss_bytes": 3773710336}, "peak_prefill_memory": null, "peak_decode_memory": 12778133536, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2e9b17f1eafffe7360614cc9ddf30c320799b427f7ef537b91abb8afc168ebe6", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}241{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "428ced74553ee05e19f78f5f", "attempt_id": "289cbb2a1598", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:08.320092Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 613, "message_tokens": 512, "message_utf8_bytes": 1542, "cache_tensor_bytes": null, "serialized_payload_bytes": 1542, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20424.734084052034, "decode_payload_ms": 0.0, "receiver_prefill_ms": 214.92629102431238, "ttft_ms": 48.71883301530033, "decode_ms": 31.92537499126047}, "total_wall_ms": 297.14354092720896, "ttft_ms": 48.71883301530033, "input_usage": {"receiver_prompt_tokens": 613, "sender_input_tokens": 1900, "sender_output_tokens": 512, "sender_prefill_ms": 491.5755409747362, "sender_ttft_ms": 46.186332940123975, "sender_decode_ms": 19884.817792102695}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3773087744, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13076094456, "mlx_cache_bytes": 6332372420, "process_rss_bytes": 3773087744}, "peak_prefill_memory": null, "peak_decode_memory": 13076094456, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ecf0d8d5109958a3a6091df20f8f76e2e6236d0c7cad9bde38d66b2ba0cfd158", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}242{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0e663e96d442d11ac6429ce0", "attempt_id": "e12529af5e91", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:08.642964Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 618, "message_tokens": 512, "message_utf8_bytes": 1542, "cache_tensor_bytes": null, "serialized_payload_bytes": 1542, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20424.734084052034, "decode_payload_ms": 0.0, "receiver_prefill_ms": 203.25029199011624, "ttft_ms": 40.22479197010398, "decode_ms": 324.21037496533245}, "total_wall_ms": 569.3384170299396, "ttft_ms": 40.22479197010398, "input_usage": {"receiver_prompt_tokens": 618, "sender_input_tokens": 1900, "sender_output_tokens": 512, "sender_prefill_ms": 491.5755409747362, "sender_ttft_ms": 46.186332940123975, "sender_decode_ms": 19884.817792102695}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 3773120512, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13075717644, "mlx_cache_bytes": 6332372420, "process_rss_bytes": 3773120512}, "peak_prefill_memory": null, "peak_decode_memory": 13075717644, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 208-A in Senneterre", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.5}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ecf0d8d5109958a3a6091df20f8f76e2e6236d0c7cad9bde38d66b2ba0cfd158", "output_hash": "8adda6ea7f7c4b4c5e25c73912f64f69b8087e33cd5ba1b20ae8ef42eb564450"}243{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "25b3cbda301f08386f38c245", "attempt_id": "0a386537cc4b", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:09.238377Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 618, "message_tokens": 512, "message_utf8_bytes": 1542, "cache_tensor_bytes": null, "serialized_payload_bytes": 1542, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20424.734084052034, "decode_payload_ms": 0.0, "receiver_prefill_ms": 182.19600000884384, "ttft_ms": 35.917041008360684, "decode_ms": 178.30633302219212}, "total_wall_ms": 397.90929097216576, "ttft_ms": 35.917041008360684, "input_usage": {"receiver_prompt_tokens": 618, "sender_input_tokens": 1900, "sender_output_tokens": 512, "sender_prefill_ms": 491.5755409747362, "sender_ttft_ms": 46.186332940123975, "sender_decode_ms": 19884.817792102695}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3773104128, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13075717644, "mlx_cache_bytes": 6332372420, "process_rss_bytes": 3773104128}, "peak_prefill_memory": null, "peak_decode_memory": 13075717644, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "54.0816", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ecf0d8d5109958a3a6091df20f8f76e2e6236d0c7cad9bde38d66b2ba0cfd158", "output_hash": "9ff09699eae63b135d2941f66a0725f6e31ea7e2b55bcd18c1c7c15a87fcce93"}244{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "1102d627ee27a9cde64cb837", "attempt_id": "193c15ce5878", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:09.659131Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 615, "message_tokens": 512, "message_utf8_bytes": 1542, "cache_tensor_bytes": null, "serialized_payload_bytes": 1542, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20424.734084052034, "decode_payload_ms": 0.0, "receiver_prefill_ms": 165.4518749564886, "ttft_ms": 26.19508292991668, "decode_ms": 183.8513750117272}, "total_wall_ms": 377.16345803346485, "ttft_ms": 26.19508292991668, "input_usage": {"receiver_prompt_tokens": 615, "sender_input_tokens": 1900, "sender_output_tokens": 512, "sender_prefill_ms": 491.5755409747362, "sender_ttft_ms": 46.186332940123975, "sender_decode_ms": 19884.817792102695}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3772956672, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13038984696, "mlx_cache_bytes": 6453089732, "process_rss_bytes": 3772956672}, "peak_prefill_memory": null, "peak_decode_memory": 13038984696, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ecf0d8d5109958a3a6091df20f8f76e2e6236d0c7cad9bde38d66b2ba0cfd158", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}245{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "97f9ba97d73aa0637beae37e", "attempt_id": "51fd52ea6708", "run_group_id": "E1_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:10.060782Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 624, "message_tokens": 512, "message_utf8_bytes": 1542, "cache_tensor_bytes": null, "serialized_payload_bytes": 1542, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20424.734084052034, "decode_payload_ms": 0.0, "receiver_prefill_ms": 286.73566703218967, "ttft_ms": 42.25175001192838, "decode_ms": 214.2483750358224}, "total_wall_ms": 544.9058330850676, "ttft_ms": 42.25175001192838, "input_usage": {"receiver_prompt_tokens": 624, "sender_input_tokens": 1900, "sender_output_tokens": 512, "sender_prefill_ms": 491.5755409747362, "sender_ttft_ms": 46.186332940123975, "sender_decode_ms": 19884.817792102695}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3772186624, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13043801636, "mlx_cache_bytes": 6767269316, "process_rss_bytes": 3772186624}, "peak_prefill_memory": null, "peak_decode_memory": 13043801636, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "ecf0d8d5109958a3a6091df20f8f76e2e6236d0c7cad9bde38d66b2ba0cfd158", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}246{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "22f80f82df58348a0e10c3e6", "attempt_id": "d66d0f8c52b6", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:10.636480Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 612, "message_tokens": 512, "message_utf8_bytes": 1513, "cache_tensor_bytes": null, "serialized_payload_bytes": 1513, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20642.006125068292, "decode_payload_ms": 0.0, "receiver_prefill_ms": 210.30787494964898, "ttft_ms": 26.419917005114257, "decode_ms": 272.1165840048343}, "total_wall_ms": 21153.805874986574, "ttft_ms": 26.419917005114257, "input_usage": {"receiver_prompt_tokens": 612, "sender_input_tokens": 1895, "sender_output_tokens": 512, "sender_prefill_ms": 528.0954170739278, "sender_ttft_ms": 46.52054200414568, "sender_decode_ms": 20062.497209059075}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 3604430848, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13075734008, "mlx_cache_bytes": 7329863108, "process_rss_bytes": 3604430848}, "peak_prefill_memory": null, "peak_decode_memory": 13075734008, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "9dd9dea09106b0487b5edd4e3fc61dc9dd37de22e7208c62a1703dc96ae635ff", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}247{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "ce981a8248ca2c2ba8607fb7", "attempt_id": "d2e851ef1855", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:31.829007Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 618, "message_tokens": 512, "message_utf8_bytes": 1486, "cache_tensor_bytes": null, "serialized_payload_bytes": 1486, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20197.262541973032, "decode_payload_ms": 0.0, "receiver_prefill_ms": 244.00791700463742, "ttft_ms": 44.42862491123378, "decode_ms": 47.71812492981553}, "total_wall_ms": 20536.538999993354, "ttft_ms": 44.42862491123378, "input_usage": {"receiver_prompt_tokens": 618, "sender_input_tokens": 1900, "sender_output_tokens": 512, "sender_prefill_ms": 554.2099579470232, "sender_ttft_ms": 30.77245899476111, "sender_decode_ms": 19609.157916042022}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3590373376, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13075717644, "mlx_cache_bytes": 7329863108, "process_rss_bytes": 3590373376}, "peak_prefill_memory": null, "peak_decode_memory": 13075717644, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "91ad43a7cddc577e740990174502e547c7cc13a83cfb2e641e88e96b3b5dd396", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}248{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a5d0a1fca4c625383e09cb20", "attempt_id": "21c9a79b74e1", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:01:52.396738Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 618, "message_tokens": 512, "message_utf8_bytes": 1474, "cache_tensor_bytes": null, "serialized_payload_bytes": 1474, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 18946.599624934606, "decode_payload_ms": 0.0, "receiver_prefill_ms": 228.98058407008648, "ttft_ms": 45.93237501103431, "decode_ms": 144.28029209375381}, "total_wall_ms": 19367.42037499789, "ttft_ms": 45.93237501103431, "input_usage": {"receiver_prompt_tokens": 618, "sender_input_tokens": 1900, "sender_output_tokens": 512, "sender_prefill_ms": 563.5605830466375, "sender_ttft_ms": 25.609166943468153, "sender_decode_ms": 18354.33445801027}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3589128192, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13075717644, "mlx_cache_bytes": 7329863108, "process_rss_bytes": 3589128192}, "peak_prefill_memory": null, "peak_decode_memory": 13075717644, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "29487d8d50f9efb64f19a47d8529d1dd58afa0ea54331d727e20ecde4accc7a1", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}249{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "4b77121bc83052f1e7672117", "attempt_id": "00e655c81d16", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:02:11.801544Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 615, "message_tokens": 512, "message_utf8_bytes": 1565, "cache_tensor_bytes": null, "serialized_payload_bytes": 1565, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20728.6467080703, "decode_payload_ms": 0.0, "receiver_prefill_ms": 187.64470797032118, "ttft_ms": 40.7695829635486, "decode_ms": 35.376082989387214}, "total_wall_ms": 20996.768583077937, "ttft_ms": 40.7695829635486, "input_usage": {"receiver_prompt_tokens": 615, "sender_input_tokens": 1897, "sender_output_tokens": 512, "sender_prefill_ms": 523.4766250941902, "sender_ttft_ms": 42.3798329429701, "sender_decode_ms": 20158.473457908258}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3584114688, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13038984696, "mlx_cache_bytes": 7516247492, "process_rss_bytes": 3584114688}, "peak_prefill_memory": null, "peak_decode_memory": 13038984696, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "2306a1c4a4df47a8f393f4be602f5a191db8cd6d5f6641e37e41a74154716155", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}250{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a411aa0cd3f1a458fe27a5fe", "attempt_id": "f60e8faafd54", "run_group_id": "E1_pilot:A4|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:02:32.827211Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 624, "message_tokens": 512, "message_utf8_bytes": 1581, "cache_tensor_bytes": null, "serialized_payload_bytes": 1581, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20795.02137505915, "decode_payload_ms": 0.0, "receiver_prefill_ms": 204.84749996103346, "ttft_ms": 36.469749989919364, "decode_ms": 197.89170799776912}, "total_wall_ms": 21236.324750003405, "ttft_ms": 36.469749989919364, "input_usage": {"receiver_prompt_tokens": 624, "sender_input_tokens": 1906, "sender_output_tokens": 512, "sender_prefill_ms": 789.7472499171272, "sender_ttft_ms": 44.0613329410553, "sender_decode_ms": 19958.223457913846}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3586129920, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13043801636, "mlx_cache_bytes": 7996486536, "process_rss_bytes": 3586129920}, "peak_prefill_memory": null, "peak_decode_memory": 13043801636, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "9b284c68fea73856904ee3e25d4ad5f7e1af8d3838510feb5dd884e88ae3768e", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}251{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6995d544152c130ebdc5d2f5", "attempt_id": "fead225c9e34", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:02:59.720083Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 189, "message_tokens": 77, "message_utf8_bytes": 210, "cache_tensor_bytes": null, "serialized_payload_bytes": 210, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5625.177541049197, "decode_payload_ms": 0.0, "receiver_prefill_ms": 97.46275003999472, "ttft_ms": 51.00066598970443, "decode_ms": 50.12999998871237}, "total_wall_ms": 199.97641700319946, "ttft_ms": 51.00066598970443, "input_usage": {"receiver_prompt_tokens": 189, "sender_input_tokens": 1930, "sender_output_tokens": 128, "sender_prefill_ms": 576.803750009276, "sender_ttft_ms": 39.65608298312873, "sender_decode_ms": 5003.769375034608}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3594797056, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12735158260, "mlx_cache_bytes": 8556876204, "process_rss_bytes": 3594797056}, "peak_prefill_memory": null, "peak_decode_memory": 12735158260, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "35dbb3ab2dc1d60cc39e141154ca09e8704d2c90fe5bb7109b5ca78ea20e71d4", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}252{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "88b52d091b388d6b7ede811e", "attempt_id": "bf48fc6f6150", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:02:59.953172Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 194, "message_tokens": 77, "message_utf8_bytes": 210, "cache_tensor_bytes": null, "serialized_payload_bytes": 210, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5625.177541049197, "decode_payload_ms": 0.0, "receiver_prefill_ms": 130.64370898064226, "ttft_ms": 61.05958390980959, "decode_ms": 35.76737502589822}, "total_wall_ms": 228.54999999981374, "ttft_ms": 61.05958390980959, "input_usage": {"receiver_prompt_tokens": 194, "sender_input_tokens": 1930, "sender_output_tokens": 128, "sender_prefill_ms": 576.803750009276, "sender_ttft_ms": 39.65608298312873, "sender_decode_ms": 5003.769375034608}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3596992512, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12738353140, "mlx_cache_bytes": 8672694700, "process_rss_bytes": 3596992512}, "peak_prefill_memory": null, "peak_decode_memory": 12738353140, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "35dbb3ab2dc1d60cc39e141154ca09e8704d2c90fe5bb7109b5ca78ea20e71d4", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}253{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a4735a7e8257ac12694fff6f", "attempt_id": "bb3e83cd2664", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:00.205379Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 194, "message_tokens": 77, "message_utf8_bytes": 210, "cache_tensor_bytes": null, "serialized_payload_bytes": 210, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5625.177541049197, "decode_payload_ms": 0.0, "receiver_prefill_ms": 81.25345897860825, "ttft_ms": 24.811208015307784, "decode_ms": 31.53608296997845}, "total_wall_ms": 138.62183294259012, "ttft_ms": 24.811208015307784, "input_usage": {"receiver_prompt_tokens": 194, "sender_input_tokens": 1930, "sender_output_tokens": 128, "sender_prefill_ms": 576.803750009276, "sender_ttft_ms": 39.65608298312873, "sender_decode_ms": 5003.769375034608}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3597139968, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12738353140, "mlx_cache_bytes": 8672694700, "process_rss_bytes": 3597139968}, "peak_prefill_memory": null, "peak_decode_memory": 12738353140, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "35dbb3ab2dc1d60cc39e141154ca09e8704d2c90fe5bb7109b5ca78ea20e71d4", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}254{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3c03e6421ed4216ad1c307b1", "attempt_id": "34f8e7cb19cb", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:00.366261Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 191, "message_tokens": 77, "message_utf8_bytes": 210, "cache_tensor_bytes": null, "serialized_payload_bytes": 210, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5625.177541049197, "decode_payload_ms": 0.0, "receiver_prefill_ms": 96.54499997850507, "ttft_ms": 38.16462494432926, "decode_ms": 21.445749909617007}, "total_wall_ms": 157.2988749248907, "ttft_ms": 38.16462494432926, "input_usage": {"receiver_prompt_tokens": 191, "sender_input_tokens": 1930, "sender_output_tokens": 128, "sender_prefill_ms": 576.803750009276, "sender_ttft_ms": 39.65608298312873, "sender_decode_ms": 5003.769375034608}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3597287424, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12746725364, "mlx_cache_bytes": 8721846700, "process_rss_bytes": 3597287424}, "peak_prefill_memory": null, "peak_decode_memory": 12746725364, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "35dbb3ab2dc1d60cc39e141154ca09e8704d2c90fe5bb7109b5ca78ea20e71d4", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}255{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "724e8d82bd118cbd836d9e9a", "attempt_id": "486254f7ade7", "run_group_id": "E1_pilot:A5|budget_tokens=128|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:00.546141Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 128, "access": "blind", "realized": {"budget_tokens": 128, "n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 200, "message_tokens": 77, "message_utf8_bytes": 210, "cache_tensor_bytes": null, "serialized_payload_bytes": 210, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5625.177541049197, "decode_payload_ms": 0.0, "receiver_prefill_ms": 113.62566601019353, "ttft_ms": 51.48070794530213, "decode_ms": 38.16312493290752}, "total_wall_ms": 204.06833302695304, "ttft_ms": 51.48070794530213, "input_usage": {"receiver_prompt_tokens": 200, "sender_input_tokens": 1930, "sender_output_tokens": 128, "sender_prefill_ms": 576.803750009276, "sender_ttft_ms": 39.65608298312873, "sender_decode_ms": 5003.769375034608}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3597385728, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12741433332, "mlx_cache_bytes": 8833470892, "process_rss_bytes": 3597385728}, "peak_prefill_memory": null, "peak_decode_memory": 12741433332, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "35dbb3ab2dc1d60cc39e141154ca09e8704d2c90fe5bb7109b5ca78ea20e71d4", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}256{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5bef07f08cb881e3c8028b8a", "attempt_id": "1ea43cead5a3", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:21.158516Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 443, "message_tokens": 331, "message_utf8_bytes": 922, "cache_tensor_bytes": null, "serialized_payload_bytes": 922, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20381.38583302498, "decode_payload_ms": 0.0, "receiver_prefill_ms": 211.09137497842312, "ttft_ms": 46.4358750032261, "decode_ms": 58.35525004658848}, "total_wall_ms": 320.40841702837497, "ttft_ms": 46.4358750032261, "input_usage": {"receiver_prompt_tokens": 443, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 542.0022499747574, "sender_ttft_ms": 51.74754199106246, "sender_decode_ms": 19785.18566605635}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3597746176, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12910582872, "mlx_cache_bytes": 9097286060, "process_rss_bytes": 3597746176}, "peak_prefill_memory": null, "peak_decode_memory": 12910582872, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e137340f68268b678688cd455e9b3cd6eb172e1f73407cc86ea551d8a3db9112", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}257{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0e23375af16b6aee689f29f3", "attempt_id": "83f8b1cc98c7", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:21.505258Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 448, "message_tokens": 331, "message_utf8_bytes": 922, "cache_tensor_bytes": null, "serialized_payload_bytes": 922, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20381.38583302498, "decode_payload_ms": 0.0, "receiver_prefill_ms": 169.4847500184551, "ttft_ms": 44.026584015227854, "decode_ms": 23.957833996973932}, "total_wall_ms": 238.91579103656113, "ttft_ms": 44.026584015227854, "input_usage": {"receiver_prompt_tokens": 448, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 542.0022499747574, "sender_ttft_ms": 51.74754199106246, "sender_decode_ms": 19785.18566605635}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3598647296, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12913482840, "mlx_cache_bytes": 9303527852, "process_rss_bytes": 3598647296}, "peak_prefill_memory": null, "peak_decode_memory": 12913482840, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e137340f68268b678688cd455e9b3cd6eb172e1f73407cc86ea551d8a3db9112", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}258{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "980b5873ccdc4bc8d0d3bea7", "attempt_id": "9568f2328455", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:21.767942Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 448, "message_tokens": 331, "message_utf8_bytes": 922, "cache_tensor_bytes": null, "serialized_payload_bytes": 922, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20381.38583302498, "decode_payload_ms": 0.0, "receiver_prefill_ms": 130.19375002477318, "ttft_ms": 39.86916702706367, "decode_ms": 165.94016598537564}, "total_wall_ms": 338.1506250007078, "ttft_ms": 39.86916702706367, "input_usage": {"receiver_prompt_tokens": 448, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 542.0022499747574, "sender_ttft_ms": 51.74754199106246, "sender_decode_ms": 19785.18566605635}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3598598144, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12913482840, "mlx_cache_bytes": 9303527852, "process_rss_bytes": 3598598144}, "peak_prefill_memory": null, "peak_decode_memory": 12913482840, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "58.0820", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e137340f68268b678688cd455e9b3cd6eb172e1f73407cc86ea551d8a3db9112", "output_hash": "b38eb8a0b5e545425ac83b4aa75ce5fda3dec19e0fb0d34757f1276e1e89f55b"}259{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "492a0ba2df92a88fd3e6ed02", "attempt_id": "4c607b027e17", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:22.158860Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 445, "message_tokens": 331, "message_utf8_bytes": 922, "cache_tensor_bytes": null, "serialized_payload_bytes": 922, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20381.38583302498, "decode_payload_ms": 0.0, "receiver_prefill_ms": 140.3645829996094, "ttft_ms": 42.18208394013345, "decode_ms": 41.18258401285857}, "total_wall_ms": 224.88687501754612, "ttft_ms": 42.18208394013345, "input_usage": {"receiver_prompt_tokens": 445, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 542.0022499747574, "sender_ttft_ms": 51.74754199106246, "sender_decode_ms": 19785.18566605635}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3598729216, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12934405208, "mlx_cache_bytes": 9461830060, "process_rss_bytes": 3598729216}, "peak_prefill_memory": null, "peak_decode_memory": 12934405208, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e137340f68268b678688cd455e9b3cd6eb172e1f73407cc86ea551d8a3db9112", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}260{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "acacd6b743285814b87f0c9e", "attempt_id": "4849490861c7", "run_group_id": "E1_pilot:A5|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:22.406409Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 454, "message_tokens": 331, "message_utf8_bytes": 922, "cache_tensor_bytes": null, "serialized_payload_bytes": 922, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20381.38583302498, "decode_payload_ms": 0.0, "receiver_prefill_ms": 120.25937507860363, "ttft_ms": 41.369040962308645, "decode_ms": 37.083958042785525}, "total_wall_ms": 200.09916694834828, "ttft_ms": 41.369040962308645, "input_usage": {"receiver_prompt_tokens": 454, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 542.0022499747574, "sender_ttft_ms": 51.74754199106246, "sender_decode_ms": 19785.18566605635}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3598942208, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12916857944, "mlx_cache_bytes": 9687470508, "process_rss_bytes": 3598942208}, "peak_prefill_memory": null, "peak_decode_memory": 12916857944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e137340f68268b678688cd455e9b3cd6eb172e1f73407cc86ea551d8a3db9112", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}261{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "6ba802baece955a4c430ccb3", "attempt_id": "a47d1fba5375", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:22.629852Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 161, "message_tokens": 50, "message_utf8_bytes": 147, "cache_tensor_bytes": null, "serialized_payload_bytes": 147, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3997.8360410314053, "decode_payload_ms": 0.0, "receiver_prefill_ms": 59.751334018073976, "ttft_ms": 28.48724997602403, "decode_ms": 224.68370792921633}, "total_wall_ms": 4311.391124967486, "ttft_ms": 28.48724997602403, "input_usage": {"receiver_prompt_tokens": 161, "sender_input_tokens": 1938, "sender_output_tokens": 87, "sender_prefill_ms": 599.2695840541273, "sender_ttft_ms": 29.5200829859823, "sender_decode_ms": 3366.715291980654}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 3599089664, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12718430196, "mlx_cache_bytes": 10271125488, "process_rss_bytes": 3599089664}, "peak_prefill_memory": null, "peak_decode_memory": 12718430196, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "c72a627e0cf29260244c44bf815ab8cf0b8705e86bc07d7006580ba6c488eda0", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}262{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5929aded800e0179e77f82b6", "attempt_id": "065d21178792", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:26.963438Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 3, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 222, "message_tokens": 105, "message_utf8_bytes": 289, "cache_tensor_bytes": null, "serialized_payload_bytes": 289, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7774.181458051316, "decode_payload_ms": 0.0, "receiver_prefill_ms": 95.50508309621364, "ttft_ms": 30.069499975070357, "decode_ms": 313.801083015278}, "total_wall_ms": 8222.901499946602, "ttft_ms": 30.069499975070357, "input_usage": {"receiver_prompt_tokens": 222, "sender_input_tokens": 1943, "sender_output_tokens": 182, "sender_prefill_ms": 599.0744999144226, "sender_ttft_ms": 48.124958062544465, "sender_decode_ms": 7124.400041997433}, "output_usage": {"output_tokens": 8}, "provider_cached_usage": null, "process_memory": 3599515648, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12763191284, "mlx_cache_bytes": 10753363528, "process_rss_bytes": 3599515648}, "peak_prefill_memory": null, "peak_decode_memory": 12763191284, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C in Schefferville", "evidence_artifact": {"n_facts": 3, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "addd40a49f09c20ba23134d69862c463ede74c1ceea507f21607a4ca03cf1ab8", "output_hash": "3491362e6ed6bb557ea2c25e29286d6fdbd0a0a36995d38446eddfd8450e097c"}263{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8af7c4e55864cbd01789515e", "attempt_id": "bde0e7a121a6", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:35.209052Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 164, "message_tokens": 48, "message_utf8_bytes": 145, "cache_tensor_bytes": null, "serialized_payload_bytes": 145, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3852.028665947728, "decode_payload_ms": 0.0, "receiver_prefill_ms": 88.19045801647007, "ttft_ms": 37.209332920610905, "decode_ms": 136.8822909425944}, "total_wall_ms": 4115.27766706422, "ttft_ms": 37.209332920610905, "input_usage": {"receiver_prompt_tokens": 164, "sender_input_tokens": 1943, "sender_output_tokens": 84, "sender_prefill_ms": 538.9621669892222, "sender_ttft_ms": 44.171374989673495, "sender_decode_ms": 3266.4305000798777}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3600252928, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12721264628, "mlx_cache_bytes": 10835283528, "process_rss_bytes": 3600252928}, "peak_prefill_memory": null, "peak_decode_memory": 12721264628, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": {"n_facts": 1, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "31baf496b39028b8cc95ba9496c01e4bb9f04e980ba244ba50daf05a4b91f94c", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}264{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f1788b1dff3714eadc8b12a0", "attempt_id": "2dc07798d547", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:39.350507Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 217, "message_tokens": 104, "message_utf8_bytes": 313, "cache_tensor_bytes": null, "serialized_payload_bytes": 313, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6098.952625063248, "decode_payload_ms": 0.0, "receiver_prefill_ms": 83.87491700705141, "ttft_ms": 47.02287493273616, "decode_ms": 206.12279104534537}, "total_wall_ms": 6437.198500032537, "ttft_ms": 47.02287493273616, "input_usage": {"receiver_prompt_tokens": 217, "sender_input_tokens": 1940, "sender_output_tokens": 142, "sender_prefill_ms": 549.4602499529719, "sender_ttft_ms": 44.020541943609715, "sender_decode_ms": 5503.327457932755}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3600859136, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12750886900, "mlx_cache_bytes": 11214261832, "process_rss_bytes": 3600859136}, "peak_prefill_memory": null, "peak_decode_memory": 12750886900, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": {"n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "0811958caf7a52851bac715e86687376bf728346117ef4252fa8c9e446fe2608", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}265{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "7c7b92fb4d30923f144f1ec7", "attempt_id": "6ffb5ede1b9a", "run_group_id": "E1_pilot:A5|budget_tokens=512|aware", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:03:45.812798Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "aware", "realized": {"budget_tokens": 512, "n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "aware", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 217, "message_tokens": 95, "message_utf8_bytes": 278, "cache_tensor_bytes": null, "serialized_payload_bytes": 278, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7845.560916000977, "decode_payload_ms": 0.0, "receiver_prefill_ms": 96.32320899982005, "ttft_ms": 36.083459039218724, "decode_ms": 172.51179204322398}, "total_wall_ms": 8151.500792009756, "ttft_ms": 36.083459039218724, "input_usage": {"receiver_prompt_tokens": 217, "sender_input_tokens": 1949, "sender_output_tokens": 182, "sender_prefill_ms": 603.8642500061542, "sender_ttft_ms": 48.5035419696942, "sender_decode_ms": 7190.499042044394}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3600728064, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 12750886900, "mlx_cache_bytes": 11705232568, "process_rss_bytes": 3600728064}, "peak_prefill_memory": null, "peak_decode_memory": 12750886900, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "7,280,585", "evidence_artifact": {"n_facts": 2, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "42dba6c4278b5de74823571ae9218b5b1c30d19c0268efeebae70b27470a211c", "output_hash": "2b81d1bd559c98218f68f4bc6bf62ababe33d07bcf6ec2cd6c2cd0508c0e52b5"}266{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "99d4df471da32d3eaa2e1485", "attempt_id": "4b1100886435", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:14.717364Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 9}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 918, "message_tokens": 821, "message_utf8_bytes": 816, "cache_tensor_bytes": null, "serialized_payload_bytes": 816, "side_information_bytes": 0, "fetched_bytes": 1496, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20721.087167039514, "decode_payload_ms": 0.0, "receiver_prefill_ms": 321.0716659668833, "ttft_ms": 42.953374912030995, "decode_ms": 230.6852920446545}, "total_wall_ms": 603.3002500189468, "ttft_ms": 42.953374912030995, "input_usage": {"receiver_prompt_tokens": 918, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 530.5115839000791, "sender_ttft_ms": 54.95133297517896, "sender_decode_ms": 20116.843333933502}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 3601760256, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13135064152, "mlx_cache_bytes": 12251032760, "process_rss_bytes": 3601760256}, "peak_prefill_memory": null, "peak_decode_memory": 13135064152, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1496, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "20e51e9edfce0f1f7bbb42564f4a801681629d693f39bdfd2f0fa18174b9872c", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}267{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "5de78be20e5e108aac62fe9c", "attempt_id": "c3fbd7f05892", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:15.378119Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 9}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 927, "message_tokens": 825, "message_utf8_bytes": 816, "cache_tensor_bytes": null, "serialized_payload_bytes": 816, "side_information_bytes": 0, "fetched_bytes": 1498, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20721.087167039514, "decode_payload_ms": 0.0, "receiver_prefill_ms": 274.8143330682069, "ttft_ms": 48.13245800323784, "decode_ms": 39.0927919652313}, "total_wall_ms": 364.51254202984273, "ttft_ms": 48.13245800323784, "input_usage": {"receiver_prompt_tokens": 927, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 530.5115839000791, "sender_ttft_ms": 54.95133297517896, "sender_decode_ms": 20116.843333933502}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3601498112, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13139323992, "mlx_cache_bytes": 12683865272, "process_rss_bytes": 3601498112}, "peak_prefill_memory": null, "peak_decode_memory": 13139323992, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1498, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "dd58c6723f7a488b688a9800bd53ab4c41b6262ba448a078f439a766616620c8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}268{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f1429aaa125142c058b71849", "attempt_id": "deacb48e5f82", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:15.770696Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 9}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 911, "message_tokens": 809, "message_utf8_bytes": 816, "cache_tensor_bytes": null, "serialized_payload_bytes": 816, "side_information_bytes": 0, "fetched_bytes": 1498, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20721.087167039514, "decode_payload_ms": 0.0, "receiver_prefill_ms": 276.44220809452236, "ttft_ms": 47.264749999158084, "decode_ms": 118.47975000273436}, "total_wall_ms": 444.5243339287117, "ttft_ms": 47.264749999158084, "input_usage": {"receiver_prompt_tokens": 911, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 530.5115839000791, "sender_ttft_ms": 54.95133297517896, "sender_decode_ms": 20116.843333933502}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3601580032, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13136096332, "mlx_cache_bytes": 13109226680, "process_rss_bytes": 3601580032}, "peak_prefill_memory": null, "peak_decode_memory": 13136096332, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1498, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "e1819d55abb157660a30ec7e8741d5ff9892523ece3b510e1752a6800ca62402", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}269{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "60fa1d3ed9e802473645af25", "attempt_id": "b17d0fc3cc79", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:16.245634Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 9}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 924, "message_tokens": 825, "message_utf8_bytes": 816, "cache_tensor_bytes": null, "serialized_payload_bytes": 816, "side_information_bytes": 0, "fetched_bytes": 1498, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20721.087167039514, "decode_payload_ms": 0.0, "receiver_prefill_ms": 254.21837496105582, "ttft_ms": 45.98754202015698, "decode_ms": 45.53999996278435}, "total_wall_ms": 347.8957919869572, "ttft_ms": 45.98754202015698, "input_usage": {"receiver_prompt_tokens": 924, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 530.5115839000791, "sender_ttft_ms": 54.95133297517896, "sender_decode_ms": 20116.843333933502}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3601645568, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13172960344, "mlx_cache_bytes": 13305932984, "process_rss_bytes": 3601645568}, "peak_prefill_memory": null, "peak_decode_memory": 13172960344, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1498, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "dd58c6723f7a488b688a9800bd53ab4c41b6262ba448a078f439a766616620c8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}270{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "a5cf7a10d76902e97c770625", "attempt_id": "d10cebc4858a", "run_group_id": "E1_pilot:A6|budget_tokens=512|max_fetch_bytes=2000|max_lookups=6|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:16.619269Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A6", "implementation_fidelity": "custom", "compression_config": {"budget_tokens": 512, "access": "blind", "max_fetch_bytes": 2000, "max_lookups": 6, "realized": {"n_refs": 9}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 933, "message_tokens": 825, "message_utf8_bytes": 816, "cache_tensor_bytes": null, "serialized_payload_bytes": 816, "side_information_bytes": 0, "fetched_bytes": 1498, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20721.087167039514, "decode_payload_ms": 0.0, "receiver_prefill_ms": 350.99987499415874, "ttft_ms": 44.614207930862904, "decode_ms": 44.20558304991573}, "total_wall_ms": 441.8494999408722, "ttft_ms": 44.614207930862904, "input_usage": {"receiver_prompt_tokens": 933, "sender_input_tokens": 1930, "sender_output_tokens": 512, "sender_prefill_ms": 530.5115839000791, "sender_ttft_ms": 54.95133297517896, "sender_decode_ms": 20116.843333933502}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3599319040, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13142240344, "mlx_cache_bytes": 13741436088, "process_rss_bytes": 3599319040}, "peak_prefill_memory": null, "peak_decode_memory": 13142240344, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"lookups": 6, "fetched_bytes": 1498, "retrieved_refs": 6}, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "dd58c6723f7a488b688a9800bd53ab4c41b6262ba448a078f439a766616620c8", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}271{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c37478f39ebe4035cc5d407f", "attempt_id": "a79639b76c38", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:17.091736Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1862, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6.017500068992376, "decode_payload_ms": 1.055374974384904, "receiver_prefill_ms": 1814.1077500768006, "ttft_ms": 76.46991696674377, "decode_ms": 210.28058300726116}, "total_wall_ms": 2104.59979204461, "ttft_ms": 76.46991696674377, "input_usage": {"receiver_prompt_tokens": 1862}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 3599433728, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13292777800, "mlx_cache_bytes": 13741436088, "process_rss_bytes": 3599433728}, "peak_prefill_memory": null, "peak_decode_memory": 13292777800, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}272{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "31002a6132e639370f1850ec", "attempt_id": "4713436e0c42", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:19.220105Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1867, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6.017500068992376, "decode_payload_ms": 1.055374974384904, "receiver_prefill_ms": 1839.8289170581847, "ttft_ms": 79.6069169882685, "decode_ms": 391.9770420761779}, "total_wall_ms": 2314.7951250430197, "ttft_ms": 79.6069169882685, "input_usage": {"receiver_prompt_tokens": 1867}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 3599876096, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13294137692, "mlx_cache_bytes": 13741436088, "process_rss_bytes": 3599876096}, "peak_prefill_memory": null, "peak_decode_memory": 13294137692, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "5280344bb130ba2ce68331d4c5deadfa47c813ecd5c007f5616e1ff44aa0daea"}273{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "cf27bcbc3d04c61c9f6e5525", "attempt_id": "2675eae23f9a", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:21.566364Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1867, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6.017500068992376, "decode_payload_ms": 1.055374974384904, "receiver_prefill_ms": 445.6370419356972, "ttft_ms": 54.01791702024639, "decode_ms": 142.66737492289394}, "total_wall_ms": 649.3898750049993, "ttft_ms": 54.01791702024639, "input_usage": {"receiver_prompt_tokens": 1867}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 3600891904, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13294137692, "mlx_cache_bytes": 13741436088, "process_rss_bytes": 3600891904}, "peak_prefill_memory": null, "peak_decode_memory": 13294137692, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "78.0424", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}274{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "d7a6807ea4e7444a7c339346", "attempt_id": "3ff2da2660f8", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:22.241231Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1864, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6.017500068992376, "decode_payload_ms": 1.055374974384904, "receiver_prefill_ms": 1245.138332946226, "ttft_ms": 39.491834002546966, "decode_ms": 205.1546670263633}, "total_wall_ms": 1494.151708087884, "ttft_ms": 39.491834002546966, "input_usage": {"receiver_prompt_tokens": 1864}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3600891904, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13350957404, "mlx_cache_bytes": 13741436088, "process_rss_bytes": 3600891904}, "peak_prefill_memory": null, "peak_decode_memory": 13350957404, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Wilhelmina Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}275{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "0498b9294db4687d717cf6b2", "attempt_id": "23d6ed0e40ac", "run_group_id": "E1_pilot:A7|codec=zstd", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:23.770962Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A7", "implementation_fidelity": "custom", "compression_config": {"codec": "zstd", "realized": {"codec": "zstd", "ratio": 3.5135135135135136}}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 1873, "message_tokens": 1766, "message_utf8_bytes": 7280, "cache_tensor_bytes": null, "serialized_payload_bytes": 2072, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 6.017500068992376, "decode_payload_ms": 1.055374974384904, "receiver_prefill_ms": 1686.4200419513509, "ttft_ms": 75.48991695512086, "decode_ms": 241.7446660110727}, "total_wall_ms": 2010.096958023496, "ttft_ms": 75.48991695512086, "input_usage": {"receiver_prompt_tokens": 1873}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 3599466496, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13295448436, "mlx_cache_bytes": 13741436088, "process_rss_bytes": 3599466496}, "peak_prefill_memory": null, "peak_decode_memory": 13295448436, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6,889,172", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "76875b24cee85c9daa41a3b2716f4617e3b85d1aa6a8a5fd34f349dcf4ae3d7e"}276{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "8c5fa58d6df871b44a252734", "attempt_id": "502dce2c3c18", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:35.953137Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.289920724801812, "origin_tokens": 1766, "compressed_tokens": 614, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 712, "message_tokens": 614, "message_utf8_bytes": 2137, "cache_tensor_bytes": null, "serialized_payload_bytes": 2137, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 10146.664374973625, "decode_payload_ms": 0.0, "receiver_prefill_ms": 2001.7235840205103, "ttft_ms": 76.0445409687236, "decode_ms": 259.1408339794725}, "total_wall_ms": 2345.732417074032, "ttft_ms": 76.0445409687236, "input_usage": {"receiver_prompt_tokens": 712}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 920567808, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13094805364, "mlx_cache_bytes": 13999536084, "process_rss_bytes": 920567808}, "peak_prefill_memory": null, "peak_decode_memory": 13094805364, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WG-14H-1182", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4ec8cdaea66fb7e8375b214400c40c867cab583421593e5329c5a55cb2af7409", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}277{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "2361486513a2aca96f995f57", "attempt_id": "ce2f5a2ceb9a", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:38.340677Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.289920724801812, "origin_tokens": 1766, "compressed_tokens": 614, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 717, "message_tokens": 614, "message_utf8_bytes": 2137, "cache_tensor_bytes": null, "serialized_payload_bytes": 2137, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 10146.664374973625, "decode_payload_ms": 0.0, "receiver_prefill_ms": 292.01187496073544, "ttft_ms": 28.167291078716516, "decode_ms": 407.5574999442324}, "total_wall_ms": 729.3605839367956, "ttft_ms": 28.167291078716516, "input_usage": {"receiver_prompt_tokens": 717}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 898138112, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13097869192, "mlx_cache_bytes": 14400209668, "process_rss_bytes": 898138112}, "peak_prefill_memory": null, "peak_decode_memory": 13097869192, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 61-C, Schefferville.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4ec8cdaea66fb7e8375b214400c40c867cab583421593e5329c5a55cb2af7409", "output_hash": "5280344bb130ba2ce68331d4c5deadfa47c813ecd5c007f5616e1ff44aa0daea"}278{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "3b990d8e10b38d4a2484cacb", "attempt_id": "01348f321875", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:39.095317Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.289920724801812, "origin_tokens": 1766, "compressed_tokens": 614, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 717, "message_tokens": 614, "message_utf8_bytes": 2137, "cache_tensor_bytes": null, "serialized_payload_bytes": 2137, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 10146.664374973625, "decode_payload_ms": 0.0, "receiver_prefill_ms": 177.63070797082037, "ttft_ms": 47.054416965693235, "decode_ms": 45.7099579507485}, "total_wall_ms": 272.4139579804614, "ttft_ms": 47.054416965693235, "input_usage": {"receiver_prompt_tokens": 717}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 898187264, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13097869192, "mlx_cache_bytes": 14400209668, "process_rss_bytes": 898187264}, "peak_prefill_memory": null, "peak_decode_memory": 13097869192, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "58", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4ec8cdaea66fb7e8375b214400c40c867cab583421593e5329c5a55cb2af7409", "output_hash": "6208ef0f7750c111548cf90b6ea1d0d0a66f6bff40dbef07cb45ec436263c7d6"}279{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "c06310f4531467fbcc43dc80", "attempt_id": "d02c96c4d959", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:39.395443Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.289920724801812, "origin_tokens": 1766, "compressed_tokens": 614, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 714, "message_tokens": 614, "message_utf8_bytes": 2137, "cache_tensor_bytes": null, "serialized_payload_bytes": 2137, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 10146.664374973625, "decode_payload_ms": 0.0, "receiver_prefill_ms": 872.7607920300215, "ttft_ms": 43.66558406036347, "decode_ms": 117.28408397175372}, "total_wall_ms": 1036.4190000109375, "ttft_ms": 43.66558406036347, "input_usage": {"receiver_prompt_tokens": 714}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 898301952, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13073801084, "mlx_cache_bytes": 14413382404, "process_rss_bytes": 898301952}, "peak_prefill_memory": null, "peak_decode_memory": 13073801084, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Haruki Rousseau", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.5}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4ec8cdaea66fb7e8375b214400c40c867cab583421593e5329c5a55cb2af7409", "output_hash": "55a01495263476a6da84eabe2c22899f74b3909040b9d9d4fe8a8b6431ff9e18"}280{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_pilot", "trial_id": "f20f0a0b88df81527aa8cf6d", "attempt_id": "8f1ce5e6206d", "run_group_id": "E1_pilot:A8|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "c7a8fcdc6928a38dfad7d8fdbe8da337078cf0a11117ad2cb6590f113531b5b2", "git_commit": "c1d4b3900e40473a307dd7a405fffb0fb01ebc3b", "timestamp_utc": "2026-10-30T04:04:40.457348Z", "task_family": "synthetic", "document_id": "syn_02000_2000", "document_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02000_2000", "sender_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "receiver_model_revision": "mlx-community/Llama-3.2-3B-Instruct-bf16@6d88ba43024fef71b10e52e101c7cd4598322601", "tokenizer_hashes": {"sender": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763", "receiver": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 12597384200}, "weight_precision": "bf16", "method_id": "A8", "implementation_fidelity": "official", "compression_config": {"budget_tokens": 512, "access": "blind", "realized": {"budget_tokens": 512, "rate": 0.289920724801812, "origin_tokens": 1766, "compressed_tokens": 614, "compressor": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank", "device": "mps"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 5, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1766, "source_tokens_receiver": 1766, "query_tokens": 723, "message_tokens": 614, "message_utf8_bytes": 2137, "cache_tensor_bytes": null, "serialized_payload_bytes": 2137, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 10146.664374973625, "decode_payload_ms": 0.0, "receiver_prefill_ms": 256.8385830381885, "ttft_ms": 41.778458980843425, "decode_ms": 174.55495800822973}, "total_wall_ms": 474.94841692969203, "ttft_ms": 41.778458980843425, "input_usage": {"receiver_prompt_tokens": 723}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 898465792, "backend_memory": {"mlx_active_bytes": 12597384200, "mlx_peak_bytes": 13101178784, "mlx_cache_bytes": 14817234508, "process_rss_bytes": 898465792}, "peak_prefill_memory": null, "peak_decode_memory": 13101178784, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "6, 889", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_pilot.log", "payload_hash": "4ec8cdaea66fb7e8375b214400c40c867cab583421593e5329c5a55cb2af7409", "output_hash": "ea1ec5a4baf1ad028e31f3f84288177e501d178ea3163a5beb1edb15fbaa34bb"}281