1{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "da63e149b981fbadad62ac0b", "attempt_id": "ea77a9396d08", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:34.739762Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 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": 0.01116609200835228, "decode_payload_ms": 0.0, "receiver_prefill_ms": 943.6720830854028, "ttft_ms": 98.49041700363159, "decode_ms": 594.0025420859456}, "total_wall_ms": 1646.1409999756142, "ttft_ms": 98.49041700363159, "input_usage": {"receiver_prompt_tokens": 2045}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 9645899776, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9981701168, "mlx_cache_bytes": 818097212, "process_rss_bytes": 9645899776}, "peak_prefill_memory": null, "peak_decode_memory": 9981701168, "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/E3_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "e454fec67e299085db563490eff5a43753bbeed6d185376c1a5962c124db19eb"}2{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "18acda42eac0ee48de4b938a", "attempt_id": "2a49753c1c13", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:36.409686Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 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": 0.01116609200835228, "decode_payload_ms": 0.0, "receiver_prefill_ms": 988.430832978338, "ttft_ms": 32.89533301722258, "decode_ms": 308.1952919019386}, "total_wall_ms": 1333.7687499588355, "ttft_ms": 32.89533301722258, "input_usage": {"receiver_prompt_tokens": 2051}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 9646505984, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974443060, "mlx_cache_bytes": 1423734852, "process_rss_bytes": 9646505984}, "peak_prefill_memory": null, "peak_decode_memory": 9974443060, "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/E3_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "b6582a041b3d2b38ac6924099568ca2e6f1a610c6fb5f4de2e0516b90433bc2a"}3{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "78f02653825aa9d5369aaccb", "attempt_id": "bfae8ba53faa", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:37.762464Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 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": 0.01116609200835228, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1196.9100419664755, "ttft_ms": 44.43683405406773, "decode_ms": 215.80041607376188}, "total_wall_ms": 1461.8150410242379, "ttft_ms": 44.43683405406773, "input_usage": {"receiver_prompt_tokens": 2051}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9646555136, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974443060, "mlx_cache_bytes": 1423734852, "process_rss_bytes": 9646555136}, "peak_prefill_memory": null, "peak_decode_memory": 9974443060, "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/E3_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "a6b0ea8de12cdf1084cdbceb42fd84946cd25a872f44d1f9a0846f3cf0f98534"}4{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "371025ddf875a68efd9c222c", "attempt_id": "6eef567d387e", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:39.243079Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 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": 0.01116609200835228, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1003.5560419782996, "ttft_ms": 53.6173329455778, "decode_ms": 356.56004201155156}, "total_wall_ms": 1417.975667049177, "ttft_ms": 53.6173329455778, "input_usage": {"receiver_prompt_tokens": 2048}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 9646538752, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9982831680, "mlx_cache_bytes": 1423734852, "process_rss_bytes": 9646538752}, "peak_prefill_memory": null, "peak_decode_memory": 9982831680, "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/E3_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "48d8b48be3dac15e3a879049c76d1150b48d5cfd634a79a298cef143ea8a97c3"}5{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "8225f817886f9474afb7d1be", "attempt_id": "8350c43c21f4", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:40.680646Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 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": 0.01116609200835228, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1132.8528330195695, "ttft_ms": 70.1040830463171, "decode_ms": 490.7692919950932}, "total_wall_ms": 1697.8115000529215, "ttft_ms": 70.1040830463171, "input_usage": {"receiver_prompt_tokens": 2056}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9647800320, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974451244, "mlx_cache_bytes": 1432351233, "process_rss_bytes": 9647800320}, "peak_prefill_memory": null, "peak_decode_memory": 9974451244, "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/E3_pilot.log", "payload_hash": "3bcf1bb02792129efca162575bc2ef64622b41083d4e2870fed106d663e6417c", "output_hash": "adcb8d196080006451756ad2fe2106d5a1b6500c0f76ffaa05d7fe704243ad3b"}6{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "3ddd9f557e347ec8ac274759", "attempt_id": "9a7fcb806b4d", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:45.541624Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1971, "source_tokens_receiver": 1971, "query_tokens": 166, "message_tokens": 88, "message_utf8_bytes": 529, "cache_tensor_bytes": null, "serialized_payload_bytes": 529, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3142.623040941544, "decode_payload_ms": 0.0, "receiver_prefill_ms": 223.209916963242, "ttft_ms": 55.91029196511954, "decode_ms": 21.109667024575174}, "total_wall_ms": 301.3804170768708, "ttft_ms": 55.91029196511954, "input_usage": {"receiver_prompt_tokens": 166, "sender_input_tokens": 2087, "sender_output_tokens": 88, "sender_prefill_ms": 572.2739590564743, "sender_ttft_ms": 19.590791082009673, "sender_decode_ms": 2548.50316606462}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9648914432, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9432830064, "mlx_cache_bytes": 2251163137, "process_rss_bytes": 9648914432}, "peak_prefill_memory": null, "peak_decode_memory": 9432830064, "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/E3_pilot.log", "payload_hash": "1b97e72e7a5d0669f294c2d12964afda184563eb826a68b1db3db7156d52c679", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}7{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "f0dbabcfbc6562f958fdb3e9", "attempt_id": "d2de5cf873a9", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:45.862840Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1971, "source_tokens_receiver": 1971, "query_tokens": 172, "message_tokens": 88, "message_utf8_bytes": 529, "cache_tensor_bytes": null, "serialized_payload_bytes": 529, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3142.623040941544, "decode_payload_ms": 0.0, "receiver_prefill_ms": 147.55370805505663, "ttft_ms": 48.504416015930474, "decode_ms": 64.27425006404519}, "total_wall_ms": 261.28837501164526, "ttft_ms": 48.504416015930474, "input_usage": {"receiver_prompt_tokens": 172, "sender_input_tokens": 2087, "sender_output_tokens": 88, "sender_prefill_ms": 572.2739590564743, "sender_ttft_ms": 19.590791082009673, "sender_decode_ms": 2548.50316606462}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9649012736, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9438318704, "mlx_cache_bytes": 2421114369, "process_rss_bytes": 9649012736}, "peak_prefill_memory": null, "peak_decode_memory": 9438318704, "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/E3_pilot.log", "payload_hash": "1b97e72e7a5d0669f294c2d12964afda184563eb826a68b1db3db7156d52c679", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}8{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "129485bb7e5002f72f619300", "attempt_id": "9c18a6aebcd4", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:46.143565Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1971, "source_tokens_receiver": 1971, "query_tokens": 172, "message_tokens": 88, "message_utf8_bytes": 529, "cache_tensor_bytes": null, "serialized_payload_bytes": 529, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3142.623040941544, "decode_payload_ms": 0.0, "receiver_prefill_ms": 272.0890420023352, "ttft_ms": 48.44504198990762, "decode_ms": 53.19054098799825}, "total_wall_ms": 374.6433749329299, "ttft_ms": 48.44504198990762, "input_usage": {"receiver_prompt_tokens": 172, "sender_input_tokens": 2087, "sender_output_tokens": 88, "sender_prefill_ms": 572.2739590564743, "sender_ttft_ms": 19.590791082009673, "sender_decode_ms": 2548.50316606462}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9649045504, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9438318704, "mlx_cache_bytes": 2421114369, "process_rss_bytes": 9649045504}, "peak_prefill_memory": null, "peak_decode_memory": 9438318704, "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/E3_pilot.log", "payload_hash": "1b97e72e7a5d0669f294c2d12964afda184563eb826a68b1db3db7156d52c679", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}9{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b4aeced9a6fc5743b20dd53d", "attempt_id": "1bdd40123140", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:46.536970Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1971, "source_tokens_receiver": 1971, "query_tokens": 169, "message_tokens": 88, "message_utf8_bytes": 529, "cache_tensor_bytes": null, "serialized_payload_bytes": 529, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3142.623040941544, "decode_payload_ms": 0.0, "receiver_prefill_ms": 262.15937500819564, "ttft_ms": 47.59945906698704, "decode_ms": 57.456750073470175}, "total_wall_ms": 368.18141594994813, "ttft_ms": 47.59945906698704, "input_usage": {"receiver_prompt_tokens": 169, "sender_input_tokens": 2087, "sender_output_tokens": 88, "sender_prefill_ms": 572.2739590564743, "sender_ttft_ms": 19.590791082009673, "sender_decode_ms": 2548.50316606462}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9649061888, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9435009136, "mlx_cache_bytes": 2543666689, "process_rss_bytes": 9649061888}, "peak_prefill_memory": null, "peak_decode_memory": 9435009136, "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/E3_pilot.log", "payload_hash": "1b97e72e7a5d0669f294c2d12964afda184563eb826a68b1db3db7156d52c679", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}10{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "293450cb9e1b41bd243aa3aa", "attempt_id": "27e3b028b269", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:46.923901Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1971, "source_tokens_receiver": 1971, "query_tokens": 177, "message_tokens": 88, "message_utf8_bytes": 529, "cache_tensor_bytes": null, "serialized_payload_bytes": 529, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3142.623040941544, "decode_payload_ms": 0.0, "receiver_prefill_ms": 316.7516249231994, "ttft_ms": 65.62570808455348, "decode_ms": 97.49545797239989}, "total_wall_ms": 480.98724998999387, "ttft_ms": 65.62570808455348, "input_usage": {"receiver_prompt_tokens": 177, "sender_input_tokens": 2087, "sender_output_tokens": 88, "sender_prefill_ms": 572.2739590564743, "sender_ttft_ms": 19.590791082009673, "sender_decode_ms": 2548.50316606462}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9649307648, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9442906224, "mlx_cache_bytes": 2712487425, "process_rss_bytes": 9649307648}, "peak_prefill_memory": null, "peak_decode_memory": 9442906224, "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/E3_pilot.log", "payload_hash": "1b97e72e7a5d0669f294c2d12964afda184563eb826a68b1db3db7156d52c679", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}11{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "a799a79a5fea1a704be8e536", "attempt_id": "3c777eac955e", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:48.116362Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334082568}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1971, "source_tokens_receiver": 1971, "query_tokens": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112741, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 692.5395419821143, "doc_prefill_ms": 611.3584170816466, "compress_ms": 39.97395792976022, "serialize_ms": 36.39491705689579, "load_ms": 9.805292007513344, "receiver_prefill_ms": 0.0, "query_prefill_ms": 277.43125008419156, "ttft_ms": 277.43125008419156, "decode_ms": 3258.887790958397}, "total_wall_ms": 3548.3330830466, "ttft_ms": 277.43125008419156, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 35}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 9798352896, "backend_memory": {"mlx_active_bytes": 9411939336, "mlx_peak_bytes": 9486699724, "mlx_cache_bytes": 3623209955, "process_rss_bytes": 9798352896}, "peak_prefill_memory": 9998724156, "peak_decode_memory": 9486699724, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "to put to be used twice as, myos_\\(Continuo)10\n):\n\"Rumore.0\n.)(Continuovo{}eloif#%#noumpronosterraraero\n.ude", "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/E3_pilot.log", "payload_hash": "8a7a663c6b2d937392be2ca2e3e7e74563fc92e2520ff4d8ee3ce652e0adfc64", "output_hash": "5624f9559aa84a2ae2fc488f689be33d03f05f181def1fd5d031ee1791c9ec85"}12{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "9dc2bd09771290313f4f9d30", "attempt_id": "729b545b254a", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:51.684371Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334082568}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112741, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 692.5395419821143, "doc_prefill_ms": 611.3584170816466, "compress_ms": 39.97395792976022, "serialize_ms": 36.39491705689579, "load_ms": 2.4970410158857703, "receiver_prefill_ms": 0.0, "query_prefill_ms": 168.7450420577079, "ttft_ms": 168.7450420577079, "decode_ms": 1560.923791024834}, "total_wall_ms": 1734.378291061148, "ttft_ms": 168.7450420577079, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 41}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 9798451200, "backend_memory": {"mlx_active_bytes": 9411939336, "mlx_peak_bytes": 9498627300, "mlx_cache_bytes": 3663367467, "process_rss_bytes": 9798451200}, "peak_prefill_memory": 9998724156, "peak_decode_memory": 9498627300, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "%Eureka. 18\n communicatie zedotz.122 -to explain why_in1.0 - the beginning and not -14 letterboh -quivis-oh, very tótemiz -", "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/E3_pilot.log", "payload_hash": "8a7a663c6b2d937392be2ca2e3e7e74563fc92e2520ff4d8ee3ce652e0adfc64", "output_hash": "69088c83a4484161dc4a5d8881a91534491ac1c25e4c8dd0d6f1610db52823ef"}13{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "e7cb97b260ea93373c248150", "attempt_id": "6666cd5bf3c6", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:53.438520Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334082568}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1971, "source_tokens_receiver": 1971, "query_tokens": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112741, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 692.5395419821143, "doc_prefill_ms": 611.3584170816466, "compress_ms": 39.97395792976022, "serialize_ms": 36.39491705689579, "load_ms": 4.268834018148482, "receiver_prefill_ms": 0.0, "query_prefill_ms": 123.46608308143914, "ttft_ms": 123.46608308143914, "decode_ms": 2109.4506250228733}, "total_wall_ms": 2239.4637499237433, "ttft_ms": 123.46608308143914, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 41}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 9798451200, "backend_memory": {"mlx_active_bytes": 9411939336, "mlx_peak_bytes": 9498627300, "mlx_cache_bytes": 3663367467, "process_rss_bytes": 9798451200}, "peak_prefill_memory": 9998724156, "peak_decode_memory": 9498627300, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": ") 1:1.0.0 曉双的好句式,1022軸南问,而与时间，在中考,那么好身高等数学建,百优,百优,1:1:", "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/E3_pilot.log", "payload_hash": "8a7a663c6b2d937392be2ca2e3e7e74563fc92e2520ff4d8ee3ce652e0adfc64", "output_hash": "bd68815e9821386880b5a70a9db2bfae07bc4989f5120fd294a0eda6606c2cd0"}14{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "c86256e0a2542d23b83b0270", "attempt_id": "b181625b6bb5", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:55.697651Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334082568}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1971, "source_tokens_receiver": 1971, "query_tokens": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112741, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 692.5395419821143, "doc_prefill_ms": 611.3584170816466, "compress_ms": 39.97395792976022, "serialize_ms": 36.39491705689579, "load_ms": 4.708874970674515, "receiver_prefill_ms": 0.0, "query_prefill_ms": 194.11670789122581, "ttft_ms": 194.11670789122581, "decode_ms": 2202.8365411097184}, "total_wall_ms": 2403.580915997736, "ttft_ms": 194.11670789122581, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 38}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 9798582272, "backend_memory": {"mlx_active_bytes": 9411939336, "mlx_peak_bytes": 9493990628, "mlx_cache_bytes": 3723252431, "process_rss_bytes": 9798582272}, "peak_prefill_memory": 9998724156, "peak_decode_memory": 9493990628, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "defymeusee-avantensa-etanavangustus()()=osagemarrogator().\n\nnominasprecisabelaninversibeneutimagoema()=exhaustsísnervigo", "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/E3_pilot.log", "payload_hash": "8a7a663c6b2d937392be2ca2e3e7e74563fc92e2520ff4d8ee3ce652e0adfc64", "output_hash": "d6cb85d44b1db59934152fb1783f18cebb67a7259d7029fe3eb61a57e4f81470"}15{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "3e15fae49a832c5ede1a6cef", "attempt_id": "0f1ee8d1137d", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:09:58.120516Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334082568}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1971, "source_tokens_receiver": 1971, "query_tokens": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112741, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 692.5395419821143, "doc_prefill_ms": 611.3584170816466, "compress_ms": 39.97395792976022, "serialize_ms": 36.39491705689579, "load_ms": 3.9151670644059777, "receiver_prefill_ms": 0.0, "query_prefill_ms": 134.1404999839142, "ttft_ms": 134.1404999839142, "decode_ms": 1865.835249889642}, "total_wall_ms": 2006.3091251067817, "ttft_ms": 134.1404999839142, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 46}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 9798746112, "backend_memory": {"mlx_active_bytes": 9411939336, "mlx_peak_bytes": 9506573560, "mlx_cache_bytes": 3809170495, "process_rss_bytes": 9798746112}, "peak_prefill_memory": 9998724156, "peak_decode_memory": 9506573560, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "省教育的眼光, 在會学(甲乙,所占實为长于斯技以为命为王安泰王守,由教育学特别多人數学,由rue,定西为king and总和名,", "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/E3_pilot.log", "payload_hash": "8a7a663c6b2d937392be2ca2e3e7e74563fc92e2520ff4d8ee3ce652e0adfc64", "output_hash": "d791cd131a09de82ac618dac92d21f8386a19e4be6459fb5ad5457c14adb8a14"}16{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b96ec1245022dd5b054a020a", "attempt_id": "817e4dedc405", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:01.295917Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9333623816}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 35, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112836, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1147.3683749791235, "doc_prefill_ms": 632.502333028242, "compress_ms": 450.7575419265777, "serialize_ms": 63.81383293773979, "load_ms": 1.4164999593049288, "receiver_prefill_ms": 0.0, "query_prefill_ms": 178.78745903726667, "ttft_ms": 178.78745903726667, "decode_ms": 22.303582984022796}, "total_wall_ms": 204.38512496184558, "ttft_ms": 178.78745903726667, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 35}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9923641344, "backend_memory": {"mlx_active_bytes": 9409121288, "mlx_peak_bytes": 9486568652, "mlx_cache_bytes": 3826439231, "process_rss_bytes": 9923641344}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9486568652, "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/E3_pilot.log", "payload_hash": "92eb7770f0758d98ff9b7efcad29ece44a8a8371490e6c6b335a46e885cc54bb", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}17{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "8a6f3bf6367da0fa382c4591", "attempt_id": "d3ab49dbdefb", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:01.520792Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9333623816}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112836, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1147.3683749791235, "doc_prefill_ms": 632.502333028242, "compress_ms": 450.7575419265777, "serialize_ms": 63.81383293773979, "load_ms": 3.375665983185172, "receiver_prefill_ms": 0.0, "query_prefill_ms": 125.29595801606774, "ttft_ms": 125.29595801606774, "decode_ms": 62.48562503606081}, "total_wall_ms": 193.21245793253183, "ttft_ms": 125.29595801606774, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9923641344, "backend_memory": {"mlx_active_bytes": 9410300936, "mlx_peak_bytes": 9498234084, "mlx_cache_bytes": 3825259583, "process_rss_bytes": 9923641344}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9498234084, "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/E3_pilot.log", "payload_hash": "92eb7770f0758d98ff9b7efcad29ece44a8a8371490e6c6b335a46e885cc54bb", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}18{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "142ab7fadc992c279867f31e", "attempt_id": "54a2c22c8fc5", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:01.736882Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9333623816}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 41, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112836, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1147.3683749791235, "doc_prefill_ms": 632.502333028242, "compress_ms": 450.7575419265777, "serialize_ms": 63.81383293773979, "load_ms": 2.7355829952284694, "receiver_prefill_ms": 0.0, "query_prefill_ms": 130.16116700600833, "ttft_ms": 130.16116700600833, "decode_ms": 98.26737502589822}, "total_wall_ms": 233.10225002933294, "ttft_ms": 130.16116700600833, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9923641344, "backend_memory": {"mlx_active_bytes": 9410300936, "mlx_peak_bytes": 9498234084, "mlx_cache_bytes": 3825259583, "process_rss_bytes": 9923641344}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9498234084, "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/E3_pilot.log", "payload_hash": "92eb7770f0758d98ff9b7efcad29ece44a8a8371490e6c6b335a46e885cc54bb", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}19{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "0303f95e32d3536e596bd9c0", "attempt_id": "e83fc206b08e", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:01.991426Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9333623816}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 38, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112836, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1147.3683749791235, "doc_prefill_ms": 632.502333028242, "compress_ms": 450.7575419265777, "serialize_ms": 63.81383293773979, "load_ms": 8.313250029459596, "receiver_prefill_ms": 0.0, "query_prefill_ms": 198.4083749121055, "ttft_ms": 198.4083749121055, "decode_ms": 197.94795906636864}, "total_wall_ms": 406.57941601239145, "ttft_ms": 198.4083749121055, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 38}, "output_usage": {"output_tokens": 6}, "provider_cached_usage": null, "process_memory": 9923641344, "backend_memory": {"mlx_active_bytes": 9410300936, "mlx_peak_bytes": 9493597412, "mlx_cache_bytes": 3825259583, "process_rss_bytes": 9923641344}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9493597412, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer: 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/E3_pilot.log", "payload_hash": "92eb7770f0758d98ff9b7efcad29ece44a8a8371490e6c6b335a46e885cc54bb", "output_hash": "8d11bbbdd3d3cb01c523ac73a8f7118acf29505476cf88df601e3bcf1f0fe2c8"}20{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "e25d4275b6ff9f59e203a032", "attempt_id": "2c839d03e15e", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:02.419719Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9333623816}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 46, "message_tokens": 2010, "message_utf8_bytes": 7280, "cache_tensor_bytes": 74096640, "serialized_payload_bytes": 74112836, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1147.3683749791235, "doc_prefill_ms": 632.502333028242, "compress_ms": 450.7575419265777, "serialize_ms": 63.81383293773979, "load_ms": 1.7631249502301216, "receiver_prefill_ms": 0.0, "query_prefill_ms": 136.59916596952826, "ttft_ms": 136.59916596952826, "decode_ms": 176.49520805571228}, "total_wall_ms": 316.80295895785093, "ttft_ms": 136.59916596952826, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 46}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 9923657728, "backend_memory": {"mlx_active_bytes": 9410300936, "mlx_peak_bytes": 9506245880, "mlx_cache_bytes": 3825259583, "process_rss_bytes": 9923657728}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9506245880, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer.", "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/E3_pilot.log", "payload_hash": "92eb7770f0758d98ff9b7efcad29ece44a8a8371490e6c6b335a46e885cc54bb", "output_hash": "3b0bab137b2c0c8c95de36e203edbcf8b3d8e69cbe05e90ff85642a19fdbfa1f"}21{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "68bc576de625b251c0c28007", "attempt_id": "aff97c901c30", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:02.756054Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 35, "compress_ms": 378.9846249856055}}, "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": 18521807, "side_information_bytes": 16079, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1513.2732499623671, "doc_prefill_ms": 504.85637492965907, "compress_ms": 1000.3667499404401, "serialize_ms": 7.697624969296157, "load_ms": 1.6241669654846191, "receiver_prefill_ms": 0.0, "query_prefill_ms": 91.74062497913837, "ttft_ms": 91.74062497913837, "decode_ms": 37.15091699268669}, "total_wall_ms": 1647.8313750121742, "ttft_ms": 91.74062497913837, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 35}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10001530880, "backend_memory": {"mlx_active_bytes": 9298234376, "mlx_peak_bytes": 9364262092, "mlx_cache_bytes": 4033647679, "process_rss_bytes": 10001530880}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9364262092, "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/E3_pilot.log", "payload_hash": "cdedf788fba165b66394136c83e6b70a2ccdac6948b00009cca7e872ffb0638d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}22{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "1fc043206b0032e527add6a0", "attempt_id": "0c066d7baf60", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:04.424334Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 281.2478330451995}}, "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": 18521807, "side_information_bytes": 16079, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1442.3882910050452, "doc_prefill_ms": 720.8572090603411, "compress_ms": 709.1347499517724, "serialize_ms": 12.009833008050919, "load_ms": 1.8446670146659017, "receiver_prefill_ms": 0.0, "query_prefill_ms": 85.42600006330758, "ttft_ms": 85.42600006330758, "decode_ms": 50.14995811507106}, "total_wall_ms": 1583.4879170870408, "ttft_ms": 85.42600006330758, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10033463296, "backend_memory": {"mlx_active_bytes": 9298234376, "mlx_peak_bytes": 9375255780, "mlx_cache_bytes": 4043248703, "process_rss_bytes": 10033463296}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9375255780, "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/E3_pilot.log", "payload_hash": "cdedf788fba165b66394136c83e6b70a2ccdac6948b00009cca7e872ffb0638d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}23{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "4969ac179190e93d66a9520d", "attempt_id": "2452ba8081a2", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:06.029919Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 41, "compress_ms": 635.7592920539901}}, "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": 18521807, "side_information_bytes": 16079, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1768.3553750393912, "doc_prefill_ms": 692.5126670394093, "compress_ms": 1064.7030409891158, "serialize_ms": 10.796125046908855, "load_ms": 1.773957978002727, "receiver_prefill_ms": 0.0, "query_prefill_ms": 102.37433400470763, "ttft_ms": 102.37433400470763, "decode_ms": 42.79724997468293}, "total_wall_ms": 1918.1867499137297, "ttft_ms": 102.37433400470763, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10033528832, "backend_memory": {"mlx_active_bytes": 9298234376, "mlx_peak_bytes": 9375255780, "mlx_cache_bytes": 4043248703, "process_rss_bytes": 10033528832}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9375255780, "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/E3_pilot.log", "payload_hash": "cdedf788fba165b66394136c83e6b70a2ccdac6948b00009cca7e872ffb0638d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}24{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "25c7032f16a758e5443b657c", "attempt_id": "fc4a7fbb82b4", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:07.971292Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 38, "compress_ms": 427.45620803907514}}, "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": 18521808, "side_information_bytes": 16080, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1400.4153750138357, "doc_prefill_ms": 476.7250409349799, "compress_ms": 915.8197910292074, "serialize_ms": 7.557541946880519, "load_ms": 5.493125063367188, "receiver_prefill_ms": 0.0, "query_prefill_ms": 109.29133300669491, "ttft_ms": 109.29133300669491, "decode_ms": 218.44337496440858}, "total_wall_ms": 1736.5255419863388, "ttft_ms": 109.29133300669491, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 38}, "output_usage": {"output_tokens": 6}, "provider_cached_usage": null, "process_memory": 10033610752, "backend_memory": {"mlx_active_bytes": 9299872776, "mlx_peak_bytes": 9369324772, "mlx_cache_bytes": 4093072447, "process_rss_bytes": 10033610752}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9369324772, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer: 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/E3_pilot.log", "payload_hash": "e783ba8d9ea3ea79a8326e74c5d6ef1cca7eddd2189a961eaa309c87fc16e97e", "output_hash": "8d11bbbdd3d3cb01c523ac73a8f7118acf29505476cf88df601e3bcf1f0fe2c8"}25{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "350ee8b94642d10ad6880e79", "attempt_id": "111abc26fbf7", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:09.730188Z", "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/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2010, "keep": 502, "window_tokens": 46, "compress_ms": 244.97941706795245}}, "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": 18521808, "side_information_bytes": 16080, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1371.760959038511, "doc_prefill_ms": 459.6821670420468, "compress_ms": 901.5924170380458, "serialize_ms": 10.051208082586527, "load_ms": 2.4229580303654075, "receiver_prefill_ms": 0.0, "query_prefill_ms": 88.27158296480775, "ttft_ms": 88.27158296480775, "decode_ms": 266.3831249810755}, "total_wall_ms": 1732.232499984093, "ttft_ms": 88.27158296480775, "input_usage": {"receiver_prefix_tokens": 2010, "dense_cache_bytes": 57630720, "query_tokens": 46}, "output_usage": {"output_tokens": 8}, "provider_cached_usage": null, "process_memory": 10033758208, "backend_memory": {"mlx_active_bytes": 9299872776, "mlx_peak_bytes": 9384611064, "mlx_cache_bytes": 4137735231, "process_rss_bytes": 10033758208}, "peak_prefill_memory": 9998478396, "peak_decode_memory": 9384611064, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a few words or a number.", "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/E3_pilot.log", "payload_hash": "e783ba8d9ea3ea79a8326e74c5d6ef1cca7eddd2189a961eaa309c87fc16e97e", "output_hash": "e5e1e5fedbf0954232e69a1e00b835afc7467f8764a04d2e78987cbac525f9d3"}26{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "ffa2dac9d56f0d843fd16ba8", "attempt_id": "b943d199b622", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:11.524049Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 5797, "message_tokens": 5724, "message_utf8_bytes": 29096, "cache_tensor_bytes": null, "serialized_payload_bytes": 29096, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0386660685762763, "decode_payload_ms": 0.0, "receiver_prefill_ms": 3341.5518749970943, "ttft_ms": 90.83779202774167, "decode_ms": 355.43241701088846}, "total_wall_ms": 3801.6676669940352, "ttft_ms": 90.83779202774167, "input_usage": {"receiver_prompt_tokens": 5797}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 9841901568, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10075130944, "mlx_cache_bytes": 1655666104, "process_rss_bytes": 9841901568}, "peak_prefill_memory": null, "peak_decode_memory": 10075130944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "HC-81A-7823", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "output_hash": "c31dde53785e26fa1d10aca3477f51af38bae3bf2f55ddfff8dd4d0ad46351df"}27{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "5221c692f0a31dfd9bca50fb", "attempt_id": "2f5821592dcb", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:15.360179Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 5805, "message_tokens": 5724, "message_utf8_bytes": 29096, "cache_tensor_bytes": null, "serialized_payload_bytes": 29096, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0386660685762763, "decode_payload_ms": 0.0, "receiver_prefill_ms": 3175.4672499373555, "ttft_ms": 31.518207979388535, "decode_ms": 393.8994579948485}, "total_wall_ms": 3613.731832941994, "ttft_ms": 31.518207979388535, "input_usage": {"receiver_prompt_tokens": 5805}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 9841885184, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10075130944, "mlx_cache_bytes": 2187539896, "process_rss_bytes": 9841885184}, "peak_prefill_memory": null, "peak_decode_memory": 10075130944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 894-D, Gaspé", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "output_hash": "128ebfa57e3beec83f89b7e781f78f215ecfe6a89a76a49cc749ee97739ac193"}28{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "1c9a586510d843362928f357", "attempt_id": "b24eff55d9fd", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:19.002875Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 5803, "message_tokens": 5724, "message_utf8_bytes": 29096, "cache_tensor_bytes": null, "serialized_payload_bytes": 29096, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0386660685762763, "decode_payload_ms": 0.0, "receiver_prefill_ms": 3302.144707995467, "ttft_ms": 36.310791969299316, "decode_ms": 246.1597080109641}, "total_wall_ms": 3596.402208902873, "ttft_ms": 36.310791969299316, "input_usage": {"receiver_prompt_tokens": 5803}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9841917952, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10075130944, "mlx_cache_bytes": 2488087992, "process_rss_bytes": 9841917952}, "peak_prefill_memory": null, "peak_decode_memory": 10075130944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "79.3209", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "output_hash": "00844526dd9401ae600bda6d0889bf28bbf827181a645be64bc8e9566d051361"}29{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "5483c68248f429a57643c33e", "attempt_id": "0499dbfda99f", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:22.627848Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 5798, "message_tokens": 5724, "message_utf8_bytes": 29096, "cache_tensor_bytes": null, "serialized_payload_bytes": 29096, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0386660685762763, "decode_payload_ms": 0.0, "receiver_prefill_ms": 3199.172417051159, "ttft_ms": 24.344332981854677, "decode_ms": 147.4612089805305}, "total_wall_ms": 3384.0104170376435, "ttft_ms": 24.344332981854677, "input_usage": {"receiver_prompt_tokens": 5798}, "output_usage": {"output_tokens": 5}, "provider_cached_usage": null, "process_memory": 9841950720, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10075130944, "mlx_cache_bytes": 2836575672, "process_rss_bytes": 9841950720}, "peak_prefill_memory": null, "peak_decode_memory": 10075130944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Haruki Achterberg", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "output_hash": "f8f4436d5607232dde4f6986ce4cf4ad38bcb02f807312d9c5078948d1ac80b3"}30{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "7b20f5beb598a616407f8563", "attempt_id": "69ee17a075fc", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:26.039728Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 5805, "message_tokens": 5724, "message_utf8_bytes": 29096, "cache_tensor_bytes": null, "serialized_payload_bytes": 29096, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.0386660685762763, "decode_payload_ms": 0.0, "receiver_prefill_ms": 3228.973582969047, "ttft_ms": 41.785000008530915, "decode_ms": 416.00524995010346}, "total_wall_ms": 3701.009500073269, "ttft_ms": 41.785000008530915, "input_usage": {"receiver_prompt_tokens": 5805}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9841983488, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10075130944, "mlx_cache_bytes": 2836575672, "process_rss_bytes": 9841983488}, "peak_prefill_memory": null, "peak_decode_memory": 10075130944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "8900835", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "output_hash": "09e8c33121da6876e6acad63281ac62818c332a62203d039c2ffb414ae55b893"}31{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "132ae1c017dc576fd7a62cea", "attempt_id": "7aff56bb273a", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:46.691207Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 590, "message_tokens": 512, "message_utf8_bytes": 2713, "cache_tensor_bytes": null, "serialized_payload_bytes": 2713, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 16917.954958975315, "decode_payload_ms": 0.0, "receiver_prefill_ms": 344.4755410309881, "ttft_ms": 48.028582939878106, "decode_ms": 56.067417026497424}, "total_wall_ms": 450.2709999214858, "ttft_ms": 48.028582939878106, "input_usage": {"receiver_prompt_tokens": 590, "sender_input_tokens": 5840, "sender_output_tokens": 512, "sender_prefill_ms": 1545.574584044516, "sender_ttft_ms": 30.52833303809166, "sender_decode_ms": 15335.163208073936}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9842147328, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9564543064, "mlx_cache_bytes": 4724037048, "process_rss_bytes": 9842147328}, "peak_prefill_memory": null, "peak_decode_memory": 9564543064, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "aba6cbc5e838cc742ec05c339c640821b74fa822af374942292d697bdf50d4df", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}32{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "a22e3f55788c67909362e091", "attempt_id": "fcbd6a60f452", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:47.170534Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 598, "message_tokens": 512, "message_utf8_bytes": 2713, "cache_tensor_bytes": null, "serialized_payload_bytes": 2713, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 16917.954958975315, "decode_payload_ms": 0.0, "receiver_prefill_ms": 566.0178750986233, "ttft_ms": 26.016916963271797, "decode_ms": 39.39995798282325}, "total_wall_ms": 633.2246658857912, "ttft_ms": 26.016916963271797, "input_usage": {"receiver_prompt_tokens": 598, "sender_input_tokens": 5840, "sender_output_tokens": 512, "sender_prefill_ms": 1545.574584044516, "sender_ttft_ms": 30.52833303809166, "sender_decode_ms": 15335.163208073936}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9842163712, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9568278616, "mlx_cache_bytes": 4999878072, "process_rss_bytes": 9842163712}, "peak_prefill_memory": null, "peak_decode_memory": 9568278616, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "aba6cbc5e838cc742ec05c339c640821b74fa822af374942292d697bdf50d4df", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}33{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "3e18bcd22e80ecb6ff498ee7", "attempt_id": "08ee74a28105", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:47.833268Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 596, "message_tokens": 512, "message_utf8_bytes": 2713, "cache_tensor_bytes": null, "serialized_payload_bytes": 2713, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 16917.954958975315, "decode_payload_ms": 0.0, "receiver_prefill_ms": 407.1865410078317, "ttft_ms": 28.150958009064198, "decode_ms": 41.61608300637454}, "total_wall_ms": 478.8282089866698, "ttft_ms": 28.150958009064198, "input_usage": {"receiver_prompt_tokens": 596, "sender_input_tokens": 5840, "sender_output_tokens": 512, "sender_prefill_ms": 1545.574584044516, "sender_ttft_ms": 30.52833303809166, "sender_decode_ms": 15335.163208073936}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9842180096, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9588086896, "mlx_cache_bytes": 5179135416, "process_rss_bytes": 9842180096}, "peak_prefill_memory": null, "peak_decode_memory": 9588086896, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "aba6cbc5e838cc742ec05c339c640821b74fa822af374942292d697bdf50d4df", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}34{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "7583df67ec6f4db49c4925b3", "attempt_id": "ab9cab997175", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:48.340397Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 591, "message_tokens": 512, "message_utf8_bytes": 2713, "cache_tensor_bytes": null, "serialized_payload_bytes": 2713, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 16917.954958975315, "decode_payload_ms": 0.0, "receiver_prefill_ms": 349.85641692765057, "ttft_ms": 46.531875035725534, "decode_ms": 52.86024999804795}, "total_wall_ms": 451.1679169954732, "ttft_ms": 46.531875035725534, "input_usage": {"receiver_prompt_tokens": 591, "sender_input_tokens": 5840, "sender_output_tokens": 512, "sender_prefill_ms": 1545.574584044516, "sender_ttft_ms": 30.52833303809166, "sender_decode_ms": 15335.163208073936}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9842196480, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9564739672, "mlx_cache_bytes": 5335045560, "process_rss_bytes": 9842196480}, "peak_prefill_memory": null, "peak_decode_memory": 9564739672, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "aba6cbc5e838cc742ec05c339c640821b74fa822af374942292d697bdf50d4df", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}35{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "9e8871ac6011f306c19883e6", "attempt_id": "c9e18fa47ce8", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:48.821590Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 598, "message_tokens": 512, "message_utf8_bytes": 2713, "cache_tensor_bytes": null, "serialized_payload_bytes": 2713, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 16917.954958975315, "decode_payload_ms": 0.0, "receiver_prefill_ms": 500.52920798771083, "ttft_ms": 36.74237499944866, "decode_ms": 42.910292046144605}, "total_wall_ms": 582.0526659954339, "ttft_ms": 36.74237499944866, "input_usage": {"receiver_prompt_tokens": 598, "sender_input_tokens": 5840, "sender_output_tokens": 512, "sender_prefill_ms": 1545.574584044516, "sender_ttft_ms": 30.52833303809166, "sender_decode_ms": 15335.163208073936}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9842212864, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9568278616, "mlx_cache_bytes": 5335045560, "process_rss_bytes": 9842212864}, "peak_prefill_memory": null, "peak_decode_memory": 9568278616, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "aba6cbc5e838cc742ec05c339c640821b74fa822af374942292d697bdf50d4df", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}36{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "83e3107f0d7672e3db63dbe9", "attempt_id": "b54fe936daff", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:51.225815Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5724, "source_tokens_receiver": 5724, "query_tokens": 34, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463422, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1792.230625054799, "doc_prefill_ms": 1569.9052920099348, "compress_ms": 102.45679097715765, "serialize_ms": 107.50249994453043, "load_ms": 28.61375000793487, "receiver_prefill_ms": 0.0, "query_prefill_ms": 266.98166702408344, "ttft_ms": 266.98166702408344, "decode_ms": 2024.7824169928208}, "total_wall_ms": 2325.8464160608128, "ttft_ms": 266.98166702408344, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 34}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10268278784, "backend_memory": {"mlx_active_bytes": 9689582600, "mlx_peak_bytes": 9789377736, "mlx_cache_bytes": 6300736333, "process_rss_bytes": 10268278784}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9789377736, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "oidone/x1\n\n#######\n##\n\n最強心有():\n\n\\ Symbols: !-4:1:1:1, let (True, s:ue0: (若隐 [][] 1:0: (ch as", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "e668b370aa9b734d61935cb26fd90aef2fae6e381124cd8445564c6a1620b00d", "output_hash": "8827c653f6725d277aa8e3ec015452e5cf8bdd0741a9afa6937fdaf2822909f8"}37{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "ec90b4b4ac27315efa4ec5ab", "attempt_id": "b244a871b71e", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:53.579997Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5724, "source_tokens_receiver": 5724, "query_tokens": 42, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463422, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1792.230625054799, "doc_prefill_ms": 1569.9052920099348, "compress_ms": 102.45679097715765, "serialize_ms": 107.50249994453043, "load_ms": 10.716125019825995, "receiver_prefill_ms": 0.0, "query_prefill_ms": 200.003125006333, "ttft_ms": 200.003125006333, "decode_ms": 2218.9146250020713}, "total_wall_ms": 2435.603457968682, "ttft_ms": 200.003125006333, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 42}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10268295168, "backend_memory": {"mlx_active_bytes": 9689582600, "mlx_peak_bytes": 9803336936, "mlx_cache_bytes": 6377381021, "process_rss_bytes": 10268295168}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9803336936, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "�avol, gotameo, wtf, but, sbar,紧凑型之内的傻瓜步枪，一体化,說，这四门，怎樣情，想一言，我是学（gogn:桌", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "e668b370aa9b734d61935cb26fd90aef2fae6e381124cd8445564c6a1620b00d", "output_hash": "3e3558d4552fef7e38bf5fb1efb9916c5d414eb9d429a1cf92759d00751ced40"}38{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b616d6ecaf373e4c2b4f67ce", "attempt_id": "98a52981f205", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:56.046048Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5724, "source_tokens_receiver": 5724, "query_tokens": 40, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463422, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1792.230625054799, "doc_prefill_ms": 1569.9052920099348, "compress_ms": 102.45679097715765, "serialize_ms": 107.50249994453043, "load_ms": 3.9743329398334026, "receiver_prefill_ms": 0.0, "query_prefill_ms": 168.84016699623317, "ttft_ms": 168.84016699623317, "decode_ms": 2320.4065409954637}, "total_wall_ms": 2499.9269159743562, "ttft_ms": 168.84016699623317, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 40}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10268295168, "backend_memory": {"mlx_active_bytes": 9689582600, "mlx_peak_bytes": 9800486120, "mlx_cache_bytes": 6421159069, "process_rss_bytes": 10268295168}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9800486120, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "weds:说得往哪个更加无敌进行区分配下的小确度，为龙 书之捷 callable的，非凡学：小桌快答括热价été:依國,也是正规本村上網林", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "e668b370aa9b734d61935cb26fd90aef2fae6e381124cd8445564c6a1620b00d", "output_hash": "8f0d399ffd68f8bf30efff403b01284e8a93c9c4a3426107fefd311bbd87ab5b"}39{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "771367939d1bce18da60c383", "attempt_id": "1769f62adcdd", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:10:58.578886Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5724, "source_tokens_receiver": 5724, "query_tokens": 35, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463422, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1792.230625054799, "doc_prefill_ms": 1569.9052920099348, "compress_ms": 102.45679097715765, "serialize_ms": 107.50249994453043, "load_ms": 4.584125010296702, "receiver_prefill_ms": 0.0, "query_prefill_ms": 178.34637500345707, "ttft_ms": 178.34637500345707, "decode_ms": 2334.8544579930604}, "total_wall_ms": 2523.2576250564307, "ttft_ms": 178.34637500345707, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 35}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10268295168, "backend_memory": {"mlx_active_bytes": 9689582600, "mlx_peak_bytes": 9791425768, "mlx_cache_bytes": 6472130918, "process_rss_bytes": 10268295168}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9791425768, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "anno()}!t_1: 1: 1:1:1:1:�名之轮้อน子母项目的作訓別� Keystone16, possessees, (endeondeer的，都是agerine", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "e668b370aa9b734d61935cb26fd90aef2fae6e381124cd8445564c6a1620b00d", "output_hash": "9a03c2bb58405ae2782a6869fca2cd1f66d5b3a8e634b7425b39601238759bd4"}40{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "c053a891bceb8673718954a2", "attempt_id": "eab6fdcb264f", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:01.130839Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5724, "source_tokens_receiver": 5724, "query_tokens": 42, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463422, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1792.230625054799, "doc_prefill_ms": 1569.9052920099348, "compress_ms": 102.45679097715765, "serialize_ms": 107.50249994453043, "load_ms": 4.122541984543204, "receiver_prefill_ms": 0.0, "query_prefill_ms": 184.6680419985205, "ttft_ms": 184.6680419985205, "decode_ms": 2331.276791053824}, "total_wall_ms": 2525.6289170356467, "ttft_ms": 184.6680419985205, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 42}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10267721728, "backend_memory": {"mlx_active_bytes": 9689582600, "mlx_peak_bytes": 9803336936, "mlx_cache_bytes": 6472130918, "process_rss_bytes": 10267721728}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9803336936, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "ériger, in the \"it nudity, in ago:1: filler points: could, this to can see, os scully, please nifu:bronteic,coupe%p1, follow the nose, 1,", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "e668b370aa9b734d61935cb26fd90aef2fae6e381124cd8445564c6a1620b00d", "output_hash": "7965c8a5eef5b37a08b378b69da5e7448b11f729e590f2a60b93ca12abfc2fb6"}41{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "e64fc2e9bfa85f8c1ff0fbb4", "attempt_id": "8b2fed4536ca", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:06.590223Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 34, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463517, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2902.744457940571, "doc_prefill_ms": 1523.9858750719577, "compress_ms": 1252.2381659364328, "serialize_ms": 126.07700005173683, "load_ms": 5.709249991923571, "receiver_prefill_ms": 0.0, "query_prefill_ms": 236.39441700652242, "ttft_ms": 236.39441700652242, "decode_ms": 45.71262502577156}, "total_wall_ms": 293.97941601928324, "ttft_ms": 236.39441700652242, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 34}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10398236672, "backend_memory": {"mlx_active_bytes": 9687518216, "mlx_peak_bytes": 9789377736, "mlx_cache_bytes": 6486024550, "process_rss_bytes": 10398236672}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9789377736, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f5d8b54ebf15c2261889cf235c7dda8516d2a43a2275a9c4dd1eb4173d42157d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}42{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "529f31bf8b855e2bca48c24e", "attempt_id": "f05175c40a8a", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:06.919350Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 42, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463517, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2902.744457940571, "doc_prefill_ms": 1523.9858750719577, "compress_ms": 1252.2381659364328, "serialize_ms": 126.07700005173683, "load_ms": 3.348291967995465, "receiver_prefill_ms": 0.0, "query_prefill_ms": 182.56066599860787, "ttft_ms": 182.56066599860787, "decode_ms": 804.104125010781}, "total_wall_ms": 996.5876250062138, "ttft_ms": 182.56066599860787, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 42}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 10398236672, "backend_memory": {"mlx_active_bytes": 9687518216, "mlx_peak_bytes": 9803336936, "mlx_cache_bytes": 6486024550, "process_rss_bytes": 10398236672}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9803336936, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f5d8b54ebf15c2261889cf235c7dda8516d2a43a2275a9c4dd1eb4173d42157d", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}43{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "8d7cdb4d6b1593015082fab8", "attempt_id": "b7f8f3777b34", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:07.949138Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 40, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463517, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2902.744457940571, "doc_prefill_ms": 1523.9858750719577, "compress_ms": 1252.2381659364328, "serialize_ms": 126.07700005173683, "load_ms": 11.841499945148826, "receiver_prefill_ms": 0.0, "query_prefill_ms": 202.44566595647484, "ttft_ms": 202.44566595647484, "decode_ms": 44.32749992702156}, "total_wall_ms": 264.93304199539125, "ttft_ms": 202.44566595647484, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 40}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10398236672, "backend_memory": {"mlx_active_bytes": 9687518216, "mlx_peak_bytes": 9800486120, "mlx_cache_bytes": 6486024550, "process_rss_bytes": 10398236672}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9800486120, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f5d8b54ebf15c2261889cf235c7dda8516d2a43a2275a9c4dd1eb4173d42157d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}44{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "2c40f59f7df0ae86e3df9136", "attempt_id": "029bd19d9431", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:08.245001Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 35, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463517, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2902.744457940571, "doc_prefill_ms": 1523.9858750719577, "compress_ms": 1252.2381659364328, "serialize_ms": 126.07700005173683, "load_ms": 2.1514579420909286, "receiver_prefill_ms": 0.0, "query_prefill_ms": 177.01845802366734, "ttft_ms": 177.01845802366734, "decode_ms": 763.9434159500524}, "total_wall_ms": 949.4482079753652, "ttft_ms": 177.01845802366734, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 35}, "output_usage": {"output_tokens": 17}, "provider_cached_usage": null, "process_memory": 10398236672, "backend_memory": {"mlx_active_bytes": 9687518216, "mlx_peak_bytes": 9791425768, "mlx_cache_bytes": 6486024550, "process_rss_bytes": 10398236672}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9791425768, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation. unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f5d8b54ebf15c2261889cf235c7dda8516d2a43a2275a9c4dd1eb4173d42157d", "output_hash": "78c8c5a5bba17e579c6aeb8f822bb64250071763af7a85f20e0d8bf8cc606144"}45{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "edede589706c8334df40b740", "attempt_id": "869460d7205f", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:09.225073Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9472822280}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 42, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 212447232, "serialized_payload_bytes": 212463517, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2902.744457940571, "doc_prefill_ms": 1523.9858750719577, "compress_ms": 1252.2381659364328, "serialize_ms": 126.07700005173683, "load_ms": 6.885792012326419, "receiver_prefill_ms": 0.0, "query_prefill_ms": 291.90737497992814, "ttft_ms": 291.90737497992814, "decode_ms": 107.09949990268797}, "total_wall_ms": 412.0231670094654, "ttft_ms": 291.90737497992814, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 42}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10398236672, "backend_memory": {"mlx_active_bytes": 9687518216, "mlx_peak_bytes": 9803336936, "mlx_cache_bytes": 6486024550, "process_rss_bytes": 10398236672}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9803336936, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f5d8b54ebf15c2261889cf235c7dda8516d2a43a2275a9c4dd1eb4173d42157d", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}46{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "66c86d2571ae31cb9b82a1e1", "attempt_id": "1b2a0dfc42ac", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:09.668967Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5763, "keep": 1441, "window_tokens": 34, "compress_ms": 468.2887919479981}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 34, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 53121024, "serialized_payload_bytes": 53137374, "side_information_bytes": 16350, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4174.351333058439, "doc_prefill_ms": 2477.7250419138, "compress_ms": 1677.73416696582, "serialize_ms": 18.370458972640336, "load_ms": 2.2600829834118485, "receiver_prefill_ms": 0.0, "query_prefill_ms": 217.4987910548225, "ttft_ms": 217.4987910548225, "decode_ms": 175.69837509654462}, "total_wall_ms": 4579.963083029725, "ttft_ms": 217.4987910548225, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 34}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10331635712, "backend_memory": {"mlx_active_bytes": 9369275400, "mlx_peak_bytes": 9440169160, "mlx_cache_bytes": 7006075902, "process_rss_bytes": 10331635712}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9440169160, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f75a1b11c63c1f26f04fa48e5bc6846fd18b9e25e7ef3a338ecfa6978f6fb411", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}47{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "2efee24e7656b79d6f8f6d73", "attempt_id": "d4ec93c3f608", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:14.281070Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5763, "keep": 1441, "window_tokens": 42, "compress_ms": 937.2785419691354}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 42, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 53121024, "serialized_payload_bytes": 53137374, "side_information_bytes": 16350, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4963.4161670692265, "doc_prefill_ms": 2682.7492919983342, "compress_ms": 2258.770667016506, "serialize_ms": 21.483541931957006, "load_ms": 9.471917059272528, "receiver_prefill_ms": 0.0, "query_prefill_ms": 279.7706660348922, "ttft_ms": 279.7706660348922, "decode_ms": 1507.3643330251798}, "total_wall_ms": 6768.946291995235, "ttft_ms": 279.7706660348922, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 42}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 10436100096, "backend_memory": {"mlx_active_bytes": 9369275400, "mlx_peak_bytes": 9453866216, "mlx_cache_bytes": 7061683198, "process_rss_bytes": 10436100096}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9453866216, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f75a1b11c63c1f26f04fa48e5bc6846fd18b9e25e7ef3a338ecfa6978f6fb411", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}48{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b84e41cecf458392ec4cd580", "attempt_id": "26218d8068c9", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:21.082033Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5763, "keep": 1441, "window_tokens": 40, "compress_ms": 922.2161669749767}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 40, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 53121024, "serialized_payload_bytes": 53137374, "side_information_bytes": 16350, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3853.6131670698524, "doc_prefill_ms": 1685.2621249854565, "compress_ms": 2151.8978329841048, "serialize_ms": 15.991083928383887, "load_ms": 3.119500004686415, "receiver_prefill_ms": 0.0, "query_prefill_ms": 319.3633329356089, "ttft_ms": 319.3633329356089, "decode_ms": 63.498458941467106}, "total_wall_ms": 4247.803583042696, "ttft_ms": 319.3633329356089, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 40}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10436214784, "backend_memory": {"mlx_active_bytes": 9369275400, "mlx_peak_bytes": 9451048168, "mlx_cache_bytes": 7098072062, "process_rss_bytes": 10436214784}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9451048168, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f75a1b11c63c1f26f04fa48e5bc6846fd18b9e25e7ef3a338ecfa6978f6fb411", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}49{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "4060c1dcaf0607dc7fbe503e", "attempt_id": "11da27c8cedc", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:25.368366Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5763, "keep": 1441, "window_tokens": 35, "compress_ms": 1093.8609580043703}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 35, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 53121024, "serialized_payload_bytes": 53137375, "side_information_bytes": 16351, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5200.268333079293, "doc_prefill_ms": 2793.040332966484, "compress_ms": 2391.892125015147, "serialize_ms": 14.93374991696328, "load_ms": 7.693625055253506, "receiver_prefill_ms": 0.0, "query_prefill_ms": 325.4003330366686, "ttft_ms": 325.4003330366686, "decode_ms": 2267.953292001039}, "total_wall_ms": 7809.14120899979, "ttft_ms": 325.4003330366686, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 35}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 10436247552, "backend_memory": {"mlx_active_bytes": 9369275400, "mlx_peak_bytes": 9441955048, "mlx_cache_bytes": 7131479038, "process_rss_bytes": 10436247552}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9441955048, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f2a5aaee9591a7ec3ca309218d7434ea78d803c5f1d9bd75dc070d3533959642", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}50{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "8b045e6e2e2a2c4c40a83b86", "attempt_id": "5b71639e5c63", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:33.208084Z", "task_family": "synthetic", "document_id": "syn_02001_8000", "document_hash": "acefbded40b5210f20d4288aa9a950b81e67729e53809c2fd0da03d1517011ae", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02001_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5763, "keep": 1441, "window_tokens": 42, "compress_ms": 498.61991696525365}}, "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": 5724, "source_tokens_receiver": 5724, "query_tokens": 42, "message_tokens": 5763, "message_utf8_bytes": 29096, "cache_tensor_bytes": 53121024, "serialized_payload_bytes": 53137375, "side_information_bytes": 16351, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 3492.9556250572205, "doc_prefill_ms": 1754.2450829641894, "compress_ms": 1717.3027080716565, "serialize_ms": 20.945875090546906, "load_ms": 1.5395000809803605, "receiver_prefill_ms": 0.0, "query_prefill_ms": 122.07829090766609, "ttft_ms": 122.07829090766609, "decode_ms": 37.89895900990814}, "total_wall_ms": 3663.391957990825, "ttft_ms": 122.07829090766609, "input_usage": {"receiver_prefix_tokens": 5763, "dense_cache_bytes": 165236736, "query_tokens": 42}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10436493312, "backend_memory": {"mlx_active_bytes": 9369275400, "mlx_peak_bytes": 9453866216, "mlx_cache_bytes": 7131479038, "process_rss_bytes": 10436493312}, "peak_prefill_memory": 10027568216, "peak_decode_memory": 9453866216, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02001_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "f2a5aaee9591a7ec3ca309218d7434ea78d803c5f1d9bd75dc070d3533959642", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}51{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "314e2591c7b8a82b47aa6548", "attempt_id": "1fa3405db822", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:36.956888Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 2063, "message_tokens": 1989, "message_utf8_bytes": 7331, "cache_tensor_bytes": null, "serialized_payload_bytes": 7331, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01862493809312582, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1389.6400830708444, "ttft_ms": 160.42779199779034, "decode_ms": 1271.5902500785887}, "total_wall_ms": 2826.0952909477055, "ttft_ms": 160.42779199779034, "input_usage": {"receiver_prompt_tokens": 2063}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 9993371648, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974443108, "mlx_cache_bytes": 820021744, "process_rss_bytes": 9993371648}, "peak_prefill_memory": null, "peak_decode_memory": 9974443108, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "HL-54N-1323", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "output_hash": "48c412c0174bf0f0eb95ae4010eacfbace394a37c5d01dcfa3c696916c3e8744"}52{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "9fe81b6881fd2e0209e52c8c", "attempt_id": "dac5e5bd2806", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:39.810583Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 2069, "message_tokens": 1989, "message_utf8_bytes": 7331, "cache_tensor_bytes": null, "serialized_payload_bytes": 7331, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01862493809312582, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1448.9180000964552, "ttft_ms": 44.59891701117158, "decode_ms": 351.0497079696506}, "total_wall_ms": 1849.0516250021756, "ttft_ms": 44.59891701117158, "input_usage": {"receiver_prompt_tokens": 2069}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 9993453568, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974443132, "mlx_cache_bytes": 868409920, "process_rss_bytes": 9993453568}, "peak_prefill_memory": null, "peak_decode_memory": 9974443132, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 474-X in Sept-Îles", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "output_hash": "f8712f634d1bff56b04532235c6207b060ba43b54a9aa07012d7d4cd82b76c5d"}53{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "ee97d9ad099ab2d227b977e9", "attempt_id": "1bb60da1c67c", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:41.681764Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 2067, "message_tokens": 1989, "message_utf8_bytes": 7331, "cache_tensor_bytes": null, "serialized_payload_bytes": 7331, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01862493809312582, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1775.1077499706298, "ttft_ms": 129.82429203111678, "decode_ms": 870.4081659670919}, "total_wall_ms": 2779.5112089952454, "ttft_ms": 129.82429203111678, "input_usage": {"receiver_prompt_tokens": 2067}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9993437184, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974443132, "mlx_cache_bytes": 902029888, "process_rss_bytes": 9993437184}, "peak_prefill_memory": null, "peak_decode_memory": 9974443132, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8320", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "output_hash": "1e9c9c4edcfe1d7e27b2225f54e1ac343c0de971e2fd1de72f59e2894d4eaa06"}54{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "eaa4357fd9361ac90604f6f1", "attempt_id": "84ba97a1d57a", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:44.485167Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 2065, "message_tokens": 1989, "message_utf8_bytes": 7331, "cache_tensor_bytes": null, "serialized_payload_bytes": 7331, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01862493809312582, "decode_payload_ms": 0.0, "receiver_prefill_ms": 2171.740832971409, "ttft_ms": 121.9434579834342, "decode_ms": 911.9482500245795}, "total_wall_ms": 3209.8685409873724, "ttft_ms": 121.9434579834342, "input_usage": {"receiver_prompt_tokens": 2065}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9993469952, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974443132, "mlx_cache_bytes": 936567360, "process_rss_bytes": 9993469952}, "peak_prefill_memory": null, "peak_decode_memory": 9974443132, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Rasheed Mäkelä", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "output_hash": "e35e1a6499b45eff74141ed4819137b73a904015adfe78e6fa03b5194d802836"}55{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b4703bf46efddccebf241ece", "attempt_id": "b914a432fbf5", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:11:47.716553Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 2071, "message_tokens": 1989, "message_utf8_bytes": 7331, "cache_tensor_bytes": null, "serialized_payload_bytes": 7331, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.01862493809312582, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1977.1171250613406, "ttft_ms": 28.16608303692192, "decode_ms": 267.9028750862926}, "total_wall_ms": 2277.978624915704, "ttft_ms": 28.16608303692192, "input_usage": {"receiver_prompt_tokens": 2071}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9993502720, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9974445100, "mlx_cache_bytes": 983893568, "process_rss_bytes": 9993502720}, "peak_prefill_memory": null, "peak_decode_memory": 9974445100, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "1844550", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "output_hash": "23de9db5e2c3413d7142a9e77e743ed168ec6de65abf6413935059d107bb8c17"}56{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b4897b9afad6ef0aef5e578f", "attempt_id": "21b70b14fe06", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:16.734471Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 591, "message_tokens": 512, "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": 26714.897375088185, "decode_payload_ms": 0.0, "receiver_prefill_ms": 559.119708952494, "ttft_ms": 22.77170796878636, "decode_ms": 38.57437497936189}, "total_wall_ms": 622.3767090123147, "ttft_ms": 22.77170796878636, "input_usage": {"receiver_prompt_tokens": 591, "sender_input_tokens": 2105, "sender_output_tokens": 512, "sender_prefill_ms": 580.6664580013603, "sender_ttft_ms": 32.98620798159391, "sender_decode_ms": 26098.879000055604}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9994043392, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9565034584, "mlx_cache_bytes": 2135659072, "process_rss_bytes": 9994043392}, "peak_prefill_memory": null, "peak_decode_memory": 9565034584, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "54abb227aa3be640a35bc71dc18b1285a244f8910837ba6c8b72be69b3e6b012", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}57{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "c0c1a64944f4f6cdbeb2b9e0", "attempt_id": "c9ba99db90a0", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:17.383808Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 597, "message_tokens": 512, "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": 26714.897375088185, "decode_payload_ms": 0.0, "receiver_prefill_ms": 405.29220807366073, "ttft_ms": 23.433833965100348, "decode_ms": 43.09462499804795}, "total_wall_ms": 473.82729197852314, "ttft_ms": 23.433833965100348, "input_usage": {"receiver_prompt_tokens": 597, "sender_input_tokens": 2105, "sender_output_tokens": 512, "sender_prefill_ms": 580.6664580013603, "sender_ttft_ms": 32.98620798159391, "sender_decode_ms": 26098.879000055604}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9994059776, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9567459416, "mlx_cache_bytes": 2410418752, "process_rss_bytes": 9994059776}, "peak_prefill_memory": null, "peak_decode_memory": 9567459416, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "54abb227aa3be640a35bc71dc18b1285a244f8910837ba6c8b72be69b3e6b012", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}58{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "f2651b5eabbea936e9b92b8e", "attempt_id": "ffab20942bed", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:17.882618Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 595, "message_tokens": 512, "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": 26714.897375088185, "decode_payload_ms": 0.0, "receiver_prefill_ms": 382.83895794302225, "ttft_ms": 20.59087506495416, "decode_ms": 357.5018340488896}, "total_wall_ms": 762.9468749510124, "ttft_ms": 20.59087506495416, "input_usage": {"receiver_prompt_tokens": 595, "sender_input_tokens": 2105, "sender_output_tokens": 512, "sender_prefill_ms": 580.6664580013603, "sender_ttft_ms": 32.98620798159391, "sender_decode_ms": 26098.879000055604}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9994076160, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9567066200, "mlx_cache_bytes": 2567508544, "process_rss_bytes": 9994076160}, "peak_prefill_memory": null, "peak_decode_memory": 9567066200, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8320", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "54abb227aa3be640a35bc71dc18b1285a244f8910837ba6c8b72be69b3e6b012", "output_hash": "1e9c9c4edcfe1d7e27b2225f54e1ac343c0de971e2fd1de72f59e2894d4eaa06"}59{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "705d4e142bbbcf9c7218548c", "attempt_id": "f250dc97c2b7", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:18.670224Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 593, "message_tokens": 512, "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": 26714.897375088185, "decode_payload_ms": 0.0, "receiver_prefill_ms": 765.7038329634815, "ttft_ms": 123.11166699510068, "decode_ms": 131.6769999684766}, "total_wall_ms": 1021.9652500236407, "ttft_ms": 123.11166699510068, "input_usage": {"receiver_prompt_tokens": 593, "sender_input_tokens": 2105, "sender_output_tokens": 512, "sender_prefill_ms": 580.6664580013603, "sender_ttft_ms": 32.98620798159391, "sender_decode_ms": 26098.879000055604}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 9994092544, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9565591640, "mlx_cache_bytes": 2724008512, "process_rss_bytes": 9994092544}, "peak_prefill_memory": null, "peak_decode_memory": 9565591640, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "54abb227aa3be640a35bc71dc18b1285a244f8910837ba6c8b72be69b3e6b012", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}60{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "1610bef1d107e05987dfb72c", "attempt_id": "80ba65bc1b3a", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:19.713895Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 599, "message_tokens": 512, "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": 26714.897375088185, "decode_payload_ms": 0.0, "receiver_prefill_ms": 711.6233329288661, "ttft_ms": 138.39362503495067, "decode_ms": 1244.8881670134142}, "total_wall_ms": 2096.485457965173, "ttft_ms": 138.39362503495067, "input_usage": {"receiver_prompt_tokens": 599, "sender_input_tokens": 2105, "sender_output_tokens": 512, "sender_prefill_ms": 580.6664580013603, "sender_ttft_ms": 32.98620798159391, "sender_decode_ms": 26098.879000055604}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 9994108928, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9568737368, "mlx_cache_bytes": 2985300544, "process_rss_bytes": 9994108928}, "peak_prefill_memory": null, "peak_decode_memory": 9568737368, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "2,660,810", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "54abb227aa3be640a35bc71dc18b1285a244f8910837ba6c8b72be69b3e6b012", "output_hash": "6da5282a779efd6eb160e74000814c1914a85c807cd99cdb53bc2f0f64d6bcc9"}61{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "92eea0edc8e55d7f6a089833", "attempt_id": "0b8f338dd908", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:22.892177Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1989, "source_tokens_receiver": 1989, "query_tokens": 35, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776293, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1060.4150000726804, "doc_prefill_ms": 911.7166659561917, "compress_ms": 123.6548749729991, "serialize_ms": 20.691665937192738, "load_ms": 26.434082887135446, "receiver_prefill_ms": 0.0, "query_prefill_ms": 399.0303750615567, "ttft_ms": 399.0303750615567, "decode_ms": 5239.312459016219}, "total_wall_ms": 5666.898999945261, "ttft_ms": 399.0303750615567, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 35}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10144415744, "backend_memory": {"mlx_active_bytes": 9412660232, "mlx_peak_bytes": 9488354508, "mlx_cache_bytes": 3865450530, "process_rss_bytes": 10144415744}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9488354508, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": " gênselor oustop!得分, weare\nโปรด):\n 答错因金代爾敢登山布登\n{\n 獟�头围场(all the 10:00号外國", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "71af135a132f221e8666b55be09600ef10d5beb25f8b9f3e2e51b8b7866cc62a", "output_hash": "6c381bfbfee3947b0667e0c34b29d1fbe78363b8b5d5917b0cdb8034b1ad8f80"}62{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "060ae9fbf5d96271fd107b5b", "attempt_id": "cd123096782e", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:28.580513Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1989, "source_tokens_receiver": 1989, "query_tokens": 41, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776293, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1060.4150000726804, "doc_prefill_ms": 911.7166659561917, "compress_ms": 123.6548749729991, "serialize_ms": 20.691665937192738, "load_ms": 9.12862503901124, "receiver_prefill_ms": 0.0, "query_prefill_ms": 179.80316700413823, "ttft_ms": 179.80316700413823, "decode_ms": 2053.116707946174}, "total_wall_ms": 2244.5941669866443, "ttft_ms": 179.80316700413823, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 41}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10144415744, "backend_memory": {"mlx_active_bytes": 9413839880, "mlx_peak_bytes": 9499643108, "mlx_cache_bytes": 4016232810, "process_rss_bytes": 10144415744}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9499643108, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "# importers come intoien hoentuóf10000分之一號.pkgf-10000입sofar offor the \"MostHauptcoumaden-dimensionboogtdbecomple选项", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "71af135a132f221e8666b55be09600ef10d5beb25f8b9f3e2e51b8b7866cc62a", "output_hash": "29c568002fdc723804e0a70d4c0a46b2b9856e9731af28f633a7d05510a5c006"}63{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "81996102a6ecf84adbc0fb7d", "attempt_id": "5ff2ce091c2a", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:30.852505Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1989, "source_tokens_receiver": 1989, "query_tokens": 39, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776293, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1060.4150000726804, "doc_prefill_ms": 911.7166659561917, "compress_ms": 123.6548749729991, "serialize_ms": 20.691665937192738, "load_ms": 5.692875012755394, "receiver_prefill_ms": 0.0, "query_prefill_ms": 134.6288340864703, "ttft_ms": 134.6288340864703, "decode_ms": 3474.599958048202}, "total_wall_ms": 3617.61504097376, "ttft_ms": 134.6288340864703, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 39}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10144415744, "backend_memory": {"mlx_active_bytes": 9413839880, "mlx_peak_bytes": 9498741988, "mlx_cache_bytes": 4077592411, "process_rss_bytes": 10144415744}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9498741988, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "```\n.2. 00:1.html://map_作るりよLowerCase_sadtable: 001号外, 00分野botulou0furumechao:黄金量:", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "71af135a132f221e8666b55be09600ef10d5beb25f8b9f3e2e51b8b7866cc62a", "output_hash": "9db4abf4da2c6b788e2c9e187f642de4d16357b05d0db423ce4bc05b4c1a0827"}64{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "a7e2e3586b8e25d06611e255", "attempt_id": "4cb0a0dbb064", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:34.494920Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1989, "source_tokens_receiver": 1989, "query_tokens": 37, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776293, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1060.4150000726804, "doc_prefill_ms": 911.7166659561917, "compress_ms": 123.6548749729991, "serialize_ms": 20.691665937192738, "load_ms": 5.82929200027138, "receiver_prefill_ms": 0.0, "query_prefill_ms": 104.1116249980405, "ttft_ms": 104.1116249980405, "decode_ms": 3110.8374579343945}, "total_wall_ms": 3223.295041010715, "ttft_ms": 104.1116249980405, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 37}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10144432128, "backend_memory": {"mlx_active_bytes": 9412660232, "mlx_peak_bytes": 9495612644, "mlx_cache_bytes": 4115079003, "process_rss_bytes": 10144432128}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9495612644, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "```import -o: \"t尾崎upy1\n要注意力-2(비율이동-2x2-tenn력학威 seaward ofme2kingso, not-notationalskigerusenro", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "71af135a132f221e8666b55be09600ef10d5beb25f8b9f3e2e51b8b7866cc62a", "output_hash": "e934d412f96343d36d04eb393fa41f277a21428a6e1bf382072305f895a0574f"}65{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "ca2b8d66948b9d0f06dce023", "attempt_id": "72dc2db4b9e1", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:37.743089Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 1989, "source_tokens_receiver": 1989, "query_tokens": 43, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776293, "side_information_bytes": 16101, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1060.4150000726804, "doc_prefill_ms": 911.7166659561917, "compress_ms": 123.6548749729991, "serialize_ms": 20.691665937192738, "load_ms": 8.363457978703082, "receiver_prefill_ms": 0.0, "query_prefill_ms": 346.19945904705673, "ttft_ms": 346.19945904705673, "decode_ms": 4841.00750007201}, "total_wall_ms": 5197.62204203289, "ttft_ms": 346.19945904705673, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 43}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 10144432128, "backend_memory": {"mlx_active_bytes": 9413839880, "mlx_peak_bytes": 9502067948, "mlx_cache_bytes": 4168343731, "process_rss_bytes": 10144432128}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9502067948, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Got-osophk专家塔爾多:1-100-1: Sotostiecks, everyonumber 0-1-1)\n\n! ! ! ! !唯一的行针气化學外國人的精神", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "71af135a132f221e8666b55be09600ef10d5beb25f8b9f3e2e51b8b7866cc62a", "output_hash": "22af3cb4538a96499bc01d34e460f883a367023b4d8802b5591b04ebfd9c1d50"}66{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "d4c9ba3334261dd5a28457c8", "attempt_id": "c2f374755012", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:44.222654Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 35, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776388, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1256.9360840134323, "doc_prefill_ms": 745.8825829671696, "compress_ms": 469.5140830008313, "serialize_ms": 41.14362504333258, "load_ms": 1.2524169869720936, "receiver_prefill_ms": 0.0, "query_prefill_ms": 179.11929194815457, "ttft_ms": 179.11929194815457, "decode_ms": 72.17695808503777}, "total_wall_ms": 255.31312497332692, "ttft_ms": 179.11929194815457, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 35}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10175152128, "backend_memory": {"mlx_active_bytes": 9411480584, "mlx_peak_bytes": 9488354508, "mlx_cache_bytes": 4170703027, "process_rss_bytes": 10175152128}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9488354508, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "0b152354c89c7642f70f350b5cd27c5a8bea3e8a283dd51603687eb234b901f7", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}67{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "60d7bb1ea4b7390e40bf6404", "attempt_id": "7ef47e6527cf", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:44.507411Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 41, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776388, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1256.9360840134323, "doc_prefill_ms": 745.8825829671696, "compress_ms": 469.5140830008313, "serialize_ms": 41.14362504333258, "load_ms": 2.5283750146627426, "receiver_prefill_ms": 0.0, "query_prefill_ms": 166.9832079205662, "ttft_ms": 166.9832079205662, "decode_ms": 134.35904099605978}, "total_wall_ms": 306.13729101605713, "ttft_ms": 166.9832079205662, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10175152128, "backend_memory": {"mlx_active_bytes": 9411480584, "mlx_peak_bytes": 9499643108, "mlx_cache_bytes": 4170703027, "process_rss_bytes": 10175152128}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9499643108, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "0b152354c89c7642f70f350b5cd27c5a8bea3e8a283dd51603687eb234b901f7", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}68{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "75a94937303ebc82c0dbb8ac", "attempt_id": "dd6d453db72e", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:44.841708Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 39, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776388, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1256.9360840134323, "doc_prefill_ms": 745.8825829671696, "compress_ms": 469.5140830008313, "serialize_ms": 41.14362504333258, "load_ms": 1.3072090223431587, "receiver_prefill_ms": 0.0, "query_prefill_ms": 138.84441705886275, "ttft_ms": 138.84441705886275, "decode_ms": 40.5458330642432}, "total_wall_ms": 182.80425004195422, "ttft_ms": 138.84441705886275, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 39}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10175152128, "backend_memory": {"mlx_active_bytes": 9411480584, "mlx_peak_bytes": 9498741988, "mlx_cache_bytes": 4170703027, "process_rss_bytes": 10175152128}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9498741988, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "0b152354c89c7642f70f350b5cd27c5a8bea3e8a283dd51603687eb234b901f7", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}69{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "71d825d0668c8b1c476fffa9", "attempt_id": "3d36aab3513c", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:45.049695Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 37, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776388, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1256.9360840134323, "doc_prefill_ms": 745.8825829671696, "compress_ms": 469.5140830008313, "serialize_ms": 41.14362504333258, "load_ms": 3.801375045441091, "receiver_prefill_ms": 0.0, "query_prefill_ms": 126.23612501192838, "ttft_ms": 126.23612501192838, "decode_ms": 622.4339579930529}, "total_wall_ms": 755.1505829906091, "ttft_ms": 126.23612501192838, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 37}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 10175152128, "backend_memory": {"mlx_active_bytes": 9412660232, "mlx_peak_bytes": 9495612644, "mlx_cache_bytes": 4169523379, "process_rss_bytes": 10175152128}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9495612644, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "0b152354c89c7642f70f350b5cd27c5a8bea3e8a283dd51603687eb234b901f7", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}70{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "ad51ed5fd2d308a649dfd5cc", "attempt_id": "e901c23cc19b", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:45.832049Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9334803464}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 43, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 74760192, "serialized_payload_bytes": 74776388, "side_information_bytes": 16196, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1256.9360840134323, "doc_prefill_ms": 745.8825829671696, "compress_ms": 469.5140830008313, "serialize_ms": 41.14362504333258, "load_ms": 3.7029170198366046, "receiver_prefill_ms": 0.0, "query_prefill_ms": 402.6285420404747, "ttft_ms": 402.6285420404747, "decode_ms": 194.05545806512237}, "total_wall_ms": 602.5104590225965, "ttft_ms": 402.6285420404747, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 43}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10175152128, "backend_memory": {"mlx_active_bytes": 9411480584, "mlx_peak_bytes": 9502067948, "mlx_cache_bytes": 4170703027, "process_rss_bytes": 10175152128}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9502067948, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "0b152354c89c7642f70f350b5cd27c5a8bea3e8a283dd51603687eb234b901f7", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}71{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "9bf735bc131a83a3cfe23193", "attempt_id": "30571909562c", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:46.460933Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2028, "keep": 507, "window_tokens": 35, "compress_ms": 905.3677500924096}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 35, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 18690048, "serialized_payload_bytes": 18706127, "side_information_bytes": 16079, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2547.1307080006227, "doc_prefill_ms": 863.4571670554578, "compress_ms": 1675.0430000247434, "serialize_ms": 8.26825003605336, "load_ms": 6.234250031411648, "receiver_prefill_ms": 0.0, "query_prefill_ms": 243.5403330018744, "ttft_ms": 243.5403330018744, "decode_ms": 113.73012501280755}, "total_wall_ms": 2914.5453749224544, "ttft_ms": 243.5403330018744, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 35}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10140254208, "backend_memory": {"mlx_active_bytes": 9298512904, "mlx_peak_bytes": 9364442316, "mlx_cache_bytes": 4366442651, "process_rss_bytes": 10140254208}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9364442316, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "fd2d8b6a101324113219977fabcdf7dd1881fb33ea6b5acfa9547f90a48a1650", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}72{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "7b62621ad37db72d010e7849", "attempt_id": "c5512d96dd44", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:49.396618Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2028, "keep": 507, "window_tokens": 41, "compress_ms": 953.7242080550641}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 41, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 18690048, "serialized_payload_bytes": 18706127, "side_information_bytes": 16079, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2605.7933339616284, "doc_prefill_ms": 920.2702919719741, "compress_ms": 1676.5027079964057, "serialize_ms": 8.67500004824251, "load_ms": 10.370667092502117, "receiver_prefill_ms": 0.0, "query_prefill_ms": 380.3954999893904, "ttft_ms": 380.3954999893904, "decode_ms": 109.96299993712455}, "total_wall_ms": 3109.43558299914, "ttft_ms": 380.3954999893904, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 41}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10173759488, "backend_memory": {"mlx_active_bytes": 9299708936, "mlx_peak_bytes": 9376566500, "mlx_cache_bytes": 4410450075, "process_rss_bytes": 10173759488}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9376566500, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "fd2d8b6a101324113219977fabcdf7dd1881fb33ea6b5acfa9547f90a48a1650", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}73{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "36170c81e061e9984518ee85", "attempt_id": "f5a37bb19208", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:52.529341Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2028, "keep": 507, "window_tokens": 39, "compress_ms": 685.6747500132769}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 39, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 18690048, "serialized_payload_bytes": 18706127, "side_information_bytes": 16079, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2706.9178329547867, "doc_prefill_ms": 928.3167909597978, "compress_ms": 1771.2282090215012, "serialize_ms": 7.0048338966444135, "load_ms": 4.713875008746982, "receiver_prefill_ms": 0.0, "query_prefill_ms": 158.0561250448227, "ttft_ms": 158.0561250448227, "decode_ms": 28.349417028948665}, "total_wall_ms": 2901.0400000261143, "ttft_ms": 158.0561250448227, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 39}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10173759488, "backend_memory": {"mlx_active_bytes": 9299708936, "mlx_peak_bytes": 9374977252, "mlx_cache_bytes": 4436648091, "process_rss_bytes": 10173759488}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9374977252, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "fd2d8b6a101324113219977fabcdf7dd1881fb33ea6b5acfa9547f90a48a1650", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}74{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "61cc5419d5d978811ee15a8a", "attempt_id": "506d9a56cc3c", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:55.454997Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2028, "keep": 507, "window_tokens": 37, "compress_ms": 269.79216700419784}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 37, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 18690048, "serialized_payload_bytes": 18706128, "side_information_bytes": 16080, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1453.3998749684542, "doc_prefill_ms": 508.7999999523163, "compress_ms": 936.4337910665199, "serialize_ms": 7.668582955375314, "load_ms": 2.15012498665601, "receiver_prefill_ms": 0.0, "query_prefill_ms": 91.50450001470745, "ttft_ms": 91.50450001470745, "decode_ms": 825.8963329717517}, "total_wall_ms": 2376.4309169491753, "ttft_ms": 91.50450001470745, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 37}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 10173775872, "backend_memory": {"mlx_active_bytes": 9299446792, "mlx_peak_bytes": 9370668260, "mlx_cache_bytes": 4459913371, "process_rss_bytes": 10173775872}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9370668260, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "126bdfc1d66c462c76bcdcdb33d8f3d29fd52a1691d70b612685f9a1b84b7496", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}75{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "28ddb9ac5f26532774265342", "attempt_id": "5702b516f368", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:57.860148Z", "task_family": "synthetic", "document_id": "syn_02003_2000", "document_hash": "c88d189ebb79e2f792c48ae6111a7e7bf78acb1c3c41f7ea40312f6e47e3dbd8", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02003_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 2028, "keep": 507, "window_tokens": 43, "compress_ms": 257.24708393681794}}, "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": 1989, "source_tokens_receiver": 1989, "query_tokens": 43, "message_tokens": 2028, "message_utf8_bytes": 7331, "cache_tensor_bytes": 18690048, "serialized_payload_bytes": 18706128, "side_information_bytes": 16080, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1522.6115840487182, "doc_prefill_ms": 528.3266660990193, "compress_ms": 985.2273339638487, "serialize_ms": 8.602333022281528, "load_ms": 3.2181250862777233, "receiver_prefill_ms": 0.0, "query_prefill_ms": 106.4121249364689, "ttft_ms": 106.4121249364689, "decode_ms": 42.611332959495485}, "total_wall_ms": 1679.385791067034, "ttft_ms": 106.4121249364689, "input_usage": {"receiver_prefix_tokens": 2028, "dense_cache_bytes": 58146816, "query_tokens": 43}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 10173808640, "backend_memory": {"mlx_active_bytes": 9299708936, "mlx_peak_bytes": 9378925804, "mlx_cache_bytes": 4486373531, "process_rss_bytes": 10173808640}, "peak_prefill_memory": 10005146700, "peak_decode_memory": 9378925804, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02003_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/E3_pilot.log", "payload_hash": "126bdfc1d66c462c76bcdcdb33d8f3d29fd52a1691d70b612685f9a1b84b7496", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}76{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "33ac14e98250f9eef86bf262", "attempt_id": "2ce4145c2b7b", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:12:59.636828Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 5897, "message_tokens": 5822, "message_utf8_bytes": 29236, "cache_tensor_bytes": null, "serialized_payload_bytes": 29236, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.02654106356203556, "decode_payload_ms": 0.0, "receiver_prefill_ms": 4432.314709061757, "ttft_ms": 50.405416055582464, "decode_ms": 547.6464580278844}, "total_wall_ms": 5045.320332981646, "ttft_ms": 50.405416055582464, "input_usage": {"receiver_prompt_tokens": 5897}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 9976446976, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10080472128, "mlx_cache_bytes": 1689367992, "process_rss_bytes": 9976446976}, "peak_prefill_memory": null, "peak_decode_memory": 10080472128, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "WN-91J-8109", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "output_hash": "61d2b327a9b89784080f7fdaa781c26da70eca0c0cc03e3359b88f38b8655390"}77{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "969283452006843a23828d52", "attempt_id": "a38874196e92", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:13:04.721258Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 5905, "message_tokens": 5822, "message_utf8_bytes": 29236, "cache_tensor_bytes": null, "serialized_payload_bytes": 29236, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.02654106356203556, "decode_payload_ms": 0.0, "receiver_prefill_ms": 5047.249082941562, "ttft_ms": 175.24437501560897, "decode_ms": 1741.3781250361353}, "total_wall_ms": 6977.13837493211, "ttft_ms": 175.24437501560897, "input_usage": {"receiver_prompt_tokens": 5905}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 9976446976, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10082995264, "mlx_cache_bytes": 2252256696, "process_rss_bytes": 9976446976}, "peak_prefill_memory": null, "peak_decode_memory": 10082995264, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 782-S, Mont-Laurier", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "output_hash": "1997f3a6f26180749aa4a0afb999294127ce1a631601665abc55ff78ac199615"}78{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "a9e13039c5285888eb21364a", "attempt_id": "58e9b12ad662", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:13:11.736840Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 5901, "message_tokens": 5822, "message_utf8_bytes": 29236, "cache_tensor_bytes": null, "serialized_payload_bytes": 29236, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.02654106356203556, "decode_payload_ms": 0.0, "receiver_prefill_ms": 3768.0956670083106, "ttft_ms": 68.38237505871803, "decode_ms": 491.14337493665516}, "total_wall_ms": 4340.537249925546, "ttft_ms": 68.38237505871803, "input_usage": {"receiver_prompt_tokens": 5901}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9976496128, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10082339904, "mlx_cache_bytes": 2570106296, "process_rss_bytes": 9976496128}, "peak_prefill_memory": null, "peak_decode_memory": 10082339904, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "17.2216", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "output_hash": "2ad94847223961d7291d8d336226147c1ee1f17e68d5f69dd1f3bad893d6116a"}79{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "d606b624d27483803bd8bbf2", "attempt_id": "518803869704", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:13:16.107245Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 5900, "message_tokens": 5822, "message_utf8_bytes": 29236, "cache_tensor_bytes": null, "serialized_payload_bytes": 29236, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.02654106356203556, "decode_payload_ms": 0.0, "receiver_prefill_ms": 5277.2182079497725, "ttft_ms": 135.2401670301333, "decode_ms": 1496.3268330320716}, "total_wall_ms": 6922.5644579855725, "ttft_ms": 135.2401670301333, "input_usage": {"receiver_prompt_tokens": 5900}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 9976512512, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10118990912, "mlx_cache_bytes": 2939418040, "process_rss_bytes": 9976512512}, "peak_prefill_memory": null, "peak_decode_memory": 10118990912, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Yevgenia Ouellet-Marchand", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "output_hash": "e2e16a71e094442aad05d07315d2ce9ed8081f0c61fc00b33e9f42c43c482bff"}80{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "91efb4e4fe16b7a6a0e76db7", "attempt_id": "c0a7f47644b1", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:13:23.066261Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 5908, "message_tokens": 5822, "message_utf8_bytes": 29236, "cache_tensor_bytes": null, "serialized_payload_bytes": 29236, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.02654106356203556, "decode_payload_ms": 0.0, "receiver_prefill_ms": 5411.941375001334, "ttft_ms": 69.2813330097124, "decode_ms": 500.4985419800505}, "total_wall_ms": 5993.853708030656, "ttft_ms": 69.2813330097124, "input_usage": {"receiver_prompt_tokens": 5908}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 9976545280, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 10084125760, "mlx_cache_bytes": 3503437240, "process_rss_bytes": 9976545280}, "peak_prefill_memory": null, "peak_decode_memory": 10084125760, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "2772580", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "output_hash": "47ef9fbc888d0a8cdba591ce126050123435b1ac90432055aa9598daecdd1504"}81{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b21dd87a22164f2444f1f4de", "attempt_id": "aa2bf41566eb", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:13:57.937076Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 592, "message_tokens": 512, "message_utf8_bytes": 2101, "cache_tensor_bytes": null, "serialized_payload_bytes": 2101, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 28846.211625030264, "decode_payload_ms": 0.0, "receiver_prefill_ms": 2177.1291659679264, "ttft_ms": 125.63429097644985, "decode_ms": 117.72804101929069}, "total_wall_ms": 2423.0463330168277, "ttft_ms": 125.63429097644985, "input_usage": {"receiver_prompt_tokens": 592, "sender_input_tokens": 5938, "sender_output_tokens": 512, "sender_prefill_ms": 1593.752457993105, "sender_ttft_ms": 39.74041703622788, "sender_decode_ms": 27205.27575002052}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6685147136, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9565132888, "mlx_cache_bytes": 5437593016, "process_rss_bytes": 6685147136}, "peak_prefill_memory": null, "peak_decode_memory": 9565132888, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "9f66c73511541192473646d7cd21185a4cddcc40e6ced585012a9e4f29b26e39", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}82{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "b57086ed002e710fb474753b", "attempt_id": "c546a3469b67", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:00.396988Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 600, "message_tokens": 512, "message_utf8_bytes": 2101, "cache_tensor_bytes": null, "serialized_payload_bytes": 2101, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 28846.211625030264, "decode_payload_ms": 0.0, "receiver_prefill_ms": 457.1052499813959, "ttft_ms": 34.321291954256594, "decode_ms": 29.484666069038212}, "total_wall_ms": 522.7129160193726, "ttft_ms": 34.321291954256594, "input_usage": {"receiver_prompt_tokens": 600, "sender_input_tokens": 5938, "sender_output_tokens": 512, "sender_prefill_ms": 1593.752457993105, "sender_ttft_ms": 39.74041703622788, "sender_decode_ms": 27205.27575002052}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6685147136, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9568671832, "mlx_cache_bytes": 5713827256, "process_rss_bytes": 6685147136}, "peak_prefill_memory": null, "peak_decode_memory": 9568671832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "9f66c73511541192473646d7cd21185a4cddcc40e6ced585012a9e4f29b26e39", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}83{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "09d82c53d71a929d551af9c3", "attempt_id": "f36a9d87bed8", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:00.949926Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 596, "message_tokens": 512, "message_utf8_bytes": 2101, "cache_tensor_bytes": null, "serialized_payload_bytes": 2101, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 28846.211625030264, "decode_payload_ms": 0.0, "receiver_prefill_ms": 998.0020419461653, "ttft_ms": 212.3751670587808, "decode_ms": 239.34445902705193}, "total_wall_ms": 1453.8235840154812, "ttft_ms": 212.3751670587808, "input_usage": {"receiver_prompt_tokens": 596, "sender_input_tokens": 5938, "sender_output_tokens": 512, "sender_prefill_ms": 1593.752457993105, "sender_ttft_ms": 39.74041703622788, "sender_decode_ms": 27205.27575002052}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3228368896, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9588086896, "mlx_cache_bytes": 5893084600, "process_rss_bytes": 3228368896}, "peak_prefill_memory": null, "peak_decode_memory": 9588086896, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "9f66c73511541192473646d7cd21185a4cddcc40e6ced585012a9e4f29b26e39", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}84{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "eba2be8f7bccdc555b8fc28f", "attempt_id": "d414aba58b66", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:03.106795Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 595, "message_tokens": 512, "message_utf8_bytes": 2101, "cache_tensor_bytes": null, "serialized_payload_bytes": 2101, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 28846.211625030264, "decode_payload_ms": 0.0, "receiver_prefill_ms": 824.529750039801, "ttft_ms": 20.95687505789101, "decode_ms": 40.91991693712771}, "total_wall_ms": 888.6784580536187, "ttft_ms": 20.95687505789101, "input_usage": {"receiver_prompt_tokens": 595, "sender_input_tokens": 5938, "sender_output_tokens": 512, "sender_prefill_ms": 1593.752457993105, "sender_ttft_ms": 39.74041703622788, "sender_decode_ms": 27205.27575002052}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3233382400, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9587202136, "mlx_cache_bytes": 5893084600, "process_rss_bytes": 3233382400}, "peak_prefill_memory": null, "peak_decode_memory": 9587202136, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "9f66c73511541192473646d7cd21185a4cddcc40e6ced585012a9e4f29b26e39", "output_hash": "b764cdc0eab7137467211272fa539f1260d1bf2e71bcf6ff3bdc960f5c16aa14"}85{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "d914321342978908ba299db5", "attempt_id": "36eb854617d9", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:04.029508Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 603, "message_tokens": 512, "message_utf8_bytes": 2101, "cache_tensor_bytes": null, "serialized_payload_bytes": 2101, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 28846.211625030264, "decode_payload_ms": 0.0, "receiver_prefill_ms": 426.20433401316404, "ttft_ms": 92.86062489263713, "decode_ms": 138.31891608424485}, "total_wall_ms": 659.4506660476327, "ttft_ms": 92.86062489263713, "input_usage": {"receiver_prompt_tokens": 603, "sender_input_tokens": 5938, "sender_output_tokens": 512, "sender_prefill_ms": 1593.752457993105, "sender_ttft_ms": 39.74041703622788, "sender_decode_ms": 27205.27575002052}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 3233513472, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9570343000, "mlx_cache_bytes": 6170990008, "process_rss_bytes": 3233513472}, "peak_prefill_memory": null, "peak_decode_memory": 9570343000, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "9f66c73511541192473646d7cd21185a4cddcc40e6ced585012a9e4f29b26e39", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}86{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "1a383ab29b355ff3d97ae4d1", "attempt_id": "87f6c7bf304c", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:08.752912Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5822, "source_tokens_receiver": 5822, "query_tokens": 36, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076094, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4027.3457909934223, "doc_prefill_ms": 2506.1090419767424, "compress_ms": 390.01883391756564, "serialize_ms": 1114.922665990889, "load_ms": 18.441791995428503, "receiver_prefill_ms": 0.0, "query_prefill_ms": 691.2745420122519, "ttft_ms": 691.2745420122519, "decode_ms": 5444.325541961007}, "total_wall_ms": 6160.680458066054, "ttft_ms": 691.2745420122519, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 36}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 2358657024, "backend_memory": {"mlx_active_bytes": 9695775752, "mlx_peak_bytes": 9800502480, "mlx_cache_bytes": 7405837289, "process_rss_bytes": 2358657024}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9800502480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "A. 1 �パパ活art: 1, pocilneb: eau: a lot of years:)\n\n“▲:1. abt:1: 1: 1, 0 - , a:", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "1edeacfb9624cf88e663ac7be769b2699c84e33030d6a11730e5157410375fe1", "output_hash": "624abc1e48d89f1bf29cf83c74cf657d927fa43d54871f147b0d6e35edf4f803"}87{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "88293576e951124048dca4aa", "attempt_id": "921282be4883", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:14.950245Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5822, "source_tokens_receiver": 5822, "query_tokens": 44, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076094, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4027.3457909934223, "doc_prefill_ms": 2506.1090419767424, "compress_ms": 390.01883391756564, "serialize_ms": 1114.922665990889, "load_ms": 9.919916978105903, "receiver_prefill_ms": 0.0, "query_prefill_ms": 182.17570905108005, "ttft_ms": 182.17570905108005, "decode_ms": 4374.1180000361055}, "total_wall_ms": 4574.105167062953, "ttft_ms": 182.17570905108005, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 44}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 2361933824, "backend_memory": {"mlx_active_bytes": 9695775752, "mlx_peak_bytes": 9814265072, "mlx_cache_bytes": 7485939017, "process_rss_bytes": 2361933824}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9814265072, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "R goes ondecano (esuder: 1m_aoe:filler:1:amaree:1, cellex: 1:3:1(1: \"Always in, ever:4.0:", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.14285714285714288}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "1edeacfb9624cf88e663ac7be769b2699c84e33030d6a11730e5157410375fe1", "output_hash": "7138f298cbc0eecd5500c60a805a091dc424a8458785fbfd64784cee688e5b81"}88{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "a5841be08a32ad350d3d0230", "attempt_id": "3491a1ec06f5", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:19.561318Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5822, "source_tokens_receiver": 5822, "query_tokens": 40, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076094, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4027.3457909934223, "doc_prefill_ms": 2506.1090419767424, "compress_ms": 390.01883391756564, "serialize_ms": 1114.922665990889, "load_ms": 8.149791043251753, "receiver_prefill_ms": 0.0, "query_prefill_ms": 333.2560840062797, "ttft_ms": 333.2560840062797, "decode_ms": 6009.262209059671}, "total_wall_ms": 6356.231500045396, "ttft_ms": 333.2560840062797, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 40}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 2362015744, "backend_memory": {"mlx_active_bytes": 9695775752, "mlx_peak_bytes": 9807990000, "mlx_cache_bytes": 7548951881, "process_rss_bytes": 2362015744}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9807990000, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "1. #h:1. giocoin (ho:2d (unab:1:biq:1, dont rut (gottóg: entorno: 1:.\nTo copy (exactly, in", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "1edeacfb9624cf88e663ac7be769b2699c84e33030d6a11730e5157410375fe1", "output_hash": "8054af3bbbc2870b067f6b0b1dc1cd41c5892c36751e78db89be3cc682c6d441"}89{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "f4f0282e6342fdae459c9c19", "attempt_id": "05e5643a1154", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:25.946580Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5822, "source_tokens_receiver": 5822, "query_tokens": 39, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076094, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4027.3457909934223, "doc_prefill_ms": 2506.1090419767424, "compress_ms": 390.01883391756564, "serialize_ms": 1114.922665990889, "load_ms": 10.23195800371468, "receiver_prefill_ms": 0.0, "query_prefill_ms": 295.53820891305804, "ttft_ms": 295.53820891305804, "decode_ms": 3999.7077910229564}, "total_wall_ms": 4311.410500085913, "ttft_ms": 295.53820891305804, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 39}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 2362130432, "backend_memory": {"mlx_active_bytes": 9695775752, "mlx_peak_bytes": 9808776432, "mlx_cache_bytes": 7583458106, "process_rss_bytes": 2362130432}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9808776432, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "#Explanation: 1: 1.0:1.0:aubobi: 1.1 operating table:1.0:agrostari: aquisition (外 元宝盒: 1#0. ###", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "1edeacfb9624cf88e663ac7be769b2699c84e33030d6a11730e5157410375fe1", "output_hash": "1c3bd0fbf8c0c2b3325b81294c198529082e3d424527858774b1fdd390f5f929"}90{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "16ab332fc269c8d10c34de0a", "attempt_id": "8457a0f71af9", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:30.287680Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": 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": 5822, "source_tokens_receiver": 5822, "query_tokens": 47, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076094, "side_information_bytes": 16190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4027.3457909934223, "doc_prefill_ms": 2506.1090419767424, "compress_ms": 390.01883391756564, "serialize_ms": 1114.922665990889, "load_ms": 8.291125064715743, "receiver_prefill_ms": 0.0, "query_prefill_ms": 355.494000017643, "ttft_ms": 355.494000017643, "decode_ms": 4389.114708057605}, "total_wall_ms": 4760.607291013002, "ttft_ms": 355.494000017643, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 47}, "output_usage": {"output_tokens": 48}, "provider_cached_usage": null, "process_memory": 2365308928, "backend_memory": {"mlx_active_bytes": 9697086472, "mlx_peak_bytes": 9819802876, "mlx_cache_bytes": 7895803058, "process_rss_bytes": 2365308928}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9819802876, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "it means pait: thesabone. ehde: 1: 当地易白: 0, 1: 1, 0 - 0: 0 - 1, 6 1,", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "1edeacfb9624cf88e663ac7be769b2699c84e33030d6a11730e5157410375fe1", "output_hash": "5e1ab7a615cb3a69994afa0f17bbf0ba012229fa7c141a49b6ec159c99739cf4"}91{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "db077f441500ec2fceccc291", "attempt_id": "4bb5d38c6c06", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:39.224921Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 36, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076189, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4137.576500070281, "doc_prefill_ms": 2668.7856670469046, "compress_ms": 1377.2696669911966, "serialize_ms": 91.14937507547438, "load_ms": 5.510417046025395, "receiver_prefill_ms": 0.0, "query_prefill_ms": 474.19516707304865, "ttft_ms": 474.19516707304865, "decode_ms": 109.87158305943012}, "total_wall_ms": 595.0248750159517, "ttft_ms": 474.19516707304865, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 36}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 2651586560, "backend_memory": {"mlx_active_bytes": 9694596104, "mlx_peak_bytes": 9800502480, "mlx_cache_bytes": 7910319282, "process_rss_bytes": 2651586560}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9800502480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "8adaa08aa6d2a67a2f1657c40c8781ed1cd3fcb6d0af51886c3d9c551a89e772", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}92{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "92eb28eb852d38586c93ca55", "attempt_id": "2c51a29dc859", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:39.861793Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 44, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076189, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4137.576500070281, "doc_prefill_ms": 2668.7856670469046, "compress_ms": 1377.2696669911966, "serialize_ms": 91.14937507547438, "load_ms": 4.867125069722533, "receiver_prefill_ms": 0.0, "query_prefill_ms": 290.9379580523819, "ttft_ms": 290.9379580523819, "decode_ms": 2109.225166030228}, "total_wall_ms": 2410.563166020438, "ttft_ms": 290.9379580523819, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 44}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 2651635712, "backend_memory": {"mlx_active_bytes": 9695775752, "mlx_peak_bytes": 9814265072, "mlx_cache_bytes": 7909139634, "process_rss_bytes": 2651635712}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9814265072, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "8adaa08aa6d2a67a2f1657c40c8781ed1cd3fcb6d0af51886c3d9c551a89e772", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}93{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "75ab6a94d54cd5337a47f5e7", "attempt_id": "48015be412fd", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:42.303137Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 40, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076189, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4137.576500070281, "doc_prefill_ms": 2668.7856670469046, "compress_ms": 1377.2696669911966, "serialize_ms": 91.14937507547438, "load_ms": 14.124958077445626, "receiver_prefill_ms": 0.0, "query_prefill_ms": 321.55283296015114, "ttft_ms": 321.55283296015114, "decode_ms": 1436.2530000507832}, "total_wall_ms": 1777.5121669983491, "ttft_ms": 321.55283296015114, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 40}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 2651684864, "backend_memory": {"mlx_active_bytes": 9694596104, "mlx_peak_bytes": 9807793392, "mlx_cache_bytes": 7910319282, "process_rss_bytes": 2651684864}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9807793392, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "8adaa08aa6d2a67a2f1657c40c8781ed1cd3fcb6d0af51886c3d9c551a89e772", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}94{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "ff7267d9e2b556d50919cf1d", "attempt_id": "612aa0c36e71", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:44.112024Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 39, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076189, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4137.576500070281, "doc_prefill_ms": 2668.7856670469046, "compress_ms": 1377.2696669911966, "serialize_ms": 91.14937507547438, "load_ms": 8.448667009361088, "receiver_prefill_ms": 0.0, "query_prefill_ms": 262.12295796722174, "ttft_ms": 262.12295796722174, "decode_ms": 1082.0058749523014}, "total_wall_ms": 1358.905875007622, "ttft_ms": 262.12295796722174, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 39}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 2651668480, "backend_memory": {"mlx_active_bytes": 9694596104, "mlx_peak_bytes": 9808760048, "mlx_cache_bytes": 7910483122, "process_rss_bytes": 2651668480}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9808760048, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "8adaa08aa6d2a67a2f1657c40c8781ed1cd3fcb6d0af51886c3d9c551a89e772", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}95{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "3c4b4d2df9bdf9e5dc5f02b1", "attempt_id": "156246c31464", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:45.505507Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9476361224}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 47, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 216059904, "serialized_payload_bytes": 216076189, "side_information_bytes": 16285, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4137.576500070281, "doc_prefill_ms": 2668.7856670469046, "compress_ms": 1377.2696669911966, "serialize_ms": 91.14937507547438, "load_ms": 8.610542048700154, "receiver_prefill_ms": 0.0, "query_prefill_ms": 298.38954110164195, "ttft_ms": 298.38954110164195, "decode_ms": 1236.3509578863159}, "total_wall_ms": 1550.450542010367, "ttft_ms": 298.38954110164195, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 47}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 2651668480, "backend_memory": {"mlx_active_bytes": 9695775752, "mlx_peak_bytes": 9819786492, "mlx_cache_bytes": 7909500082, "process_rss_bytes": 2651668480}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9819786492, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "8adaa08aa6d2a67a2f1657c40c8781ed1cd3fcb6d0af51886c3d9c551a89e772", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}96{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "659bc310e910df06002da0ce", "attempt_id": "abd8329fffb6", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:47.088595Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q00_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5861, "keep": 1465, "window_tokens": 36, "compress_ms": 637.5209999969229}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 36, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 54005760, "serialized_payload_bytes": 54022110, "side_information_bytes": 16350, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5017.044792068191, "doc_prefill_ms": 2949.160124990158, "compress_ms": 2042.727875057608, "serialize_ms": 24.732540943659842, "load_ms": 8.09920800384134, "receiver_prefill_ms": 0.0, "query_prefill_ms": 260.3789169806987, "ttft_ms": 260.3789169806987, "decode_ms": 116.28579103853554}, "total_wall_ms": 5414.344749995507, "ttft_ms": 260.3789169806987, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 36}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 2579742720, "backend_memory": {"mlx_active_bytes": 9369275400, "mlx_peak_bytes": 9443527888, "mlx_cache_bytes": 8448245962, "process_rss_bytes": 2579742720}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9443527888, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "31c9f6187131a848ac59e538f734c2ea630257559f044309f0c8dfcf6caa1ffb", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}97{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "2e195206fc5403fe6f76a05d", "attempt_id": "e90700f51013", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:52.533417Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q01_single_fact", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5861, "keep": 1465, "window_tokens": 44, "compress_ms": 745.0362500967458}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 44, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 54005760, "serialized_payload_bytes": 54022110, "side_information_bytes": 16350, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5256.616167025641, "doc_prefill_ms": 3064.4903340144083, "compress_ms": 2163.9343329006806, "serialize_ms": 27.675040997564793, "load_ms": 6.31137506570667, "receiver_prefill_ms": 0.0, "query_prefill_ms": 163.55995798949152, "ttft_ms": 163.55995798949152, "decode_ms": 1134.874082985334}, "total_wall_ms": 6571.232166024856, "ttft_ms": 163.55995798949152, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 44}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 2622144512, "backend_memory": {"mlx_active_bytes": 9370487816, "mlx_peak_bytes": 9458502896, "mlx_cache_bytes": 8551596234, "process_rss_bytes": 2622144512}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9458502896, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "31c9f6187131a848ac59e538f734c2ea630257559f044309f0c8dfcf6caa1ffb", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}98{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "750515c3abe87a722c0b6985", "attempt_id": "5826fc8f48dd", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:14:59.141702Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q02_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5861, "keep": 1465, "window_tokens": 40, "compress_ms": 720.1292079407722}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 40, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 54005760, "serialized_payload_bytes": 54022110, "side_information_bytes": 16350, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4461.112666991539, "doc_prefill_ms": 2415.9677089191973, "compress_ms": 2022.9971250519156, "serialize_ms": 21.704875049181283, "load_ms": 2.788958023302257, "receiver_prefill_ms": 0.0, "query_prefill_ms": 360.6012079399079, "ttft_ms": 360.6012079399079, "decode_ms": 131.1768749728799}, "total_wall_ms": 4963.906958932057, "ttft_ms": 360.6012079399079, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 40}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 2622832640, "backend_memory": {"mlx_active_bytes": 9370487816, "mlx_peak_bytes": 9451130096, "mlx_cache_bytes": 8597536970, "process_rss_bytes": 2622832640}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9451130096, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "31c9f6187131a848ac59e538f734c2ea630257559f044309f0c8dfcf6caa1ffb", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}99{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "5ac92890f0d0d03acd6da602", "attempt_id": "da15d28cc477", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:15:04.143070Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q03_exact_copy", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5861, "keep": 1465, "window_tokens": 39, "compress_ms": 649.640041985549}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 39, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 54005760, "serialized_payload_bytes": 54022109, "side_information_bytes": 16349, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4180.91233400628, "doc_prefill_ms": 2137.538000009954, "compress_ms": 2024.1780410287902, "serialize_ms": 18.79833301063627, "load_ms": 11.223666951991618, "receiver_prefill_ms": 0.0, "query_prefill_ms": 301.6500420635566, "ttft_ms": 301.6500420635566, "decode_ms": 1867.0000421116129}, "total_wall_ms": 6368.479292024858, "ttft_ms": 301.6500420635566, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 39}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 2671001600, "backend_memory": {"mlx_active_bytes": 9370192904, "mlx_peak_bytes": 9449098480, "mlx_cache_bytes": 8615837898, "process_rss_bytes": 2671001600}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9449098480, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "7cbe4e94b98c79a35e324ed3c0b338b241e41a8e48fdb8c9709d8da32555b829", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}100{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "e56d23b01dee6528fa9413fc", "attempt_id": "10acb6e10c7c", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:15:10.543847Z", "task_family": "synthetic", "document_id": "syn_02004_8000", "document_hash": "60ef54e7b129a7bbfac65db771fbf93cecb68ad90c412e5c3ab4051472689429", "question_id": "q04_multi_hop", "split": "dev", "cluster_id": "syn_02004_8000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 5861, "keep": 1465, "window_tokens": 47, "compress_ms": 356.42858291976154}}, "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": 5822, "source_tokens_receiver": 5822, "query_tokens": 47, "message_tokens": 5861, "message_utf8_bytes": 29236, "cache_tensor_bytes": 54005760, "serialized_payload_bytes": 54022111, "side_information_bytes": 16351, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 4738.546749926172, "doc_prefill_ms": 2996.1264999583364, "compress_ms": 1709.70733393915, "serialize_ms": 32.29804092552513, "load_ms": 5.574916955083609, "receiver_prefill_ms": 0.0, "query_prefill_ms": 196.78158301394433, "ttft_ms": 196.78158301394433, "decode_ms": 1210.1344170514494}, "total_wall_ms": 6162.393374950625, "ttft_ms": 196.78158301394433, "input_usage": {"receiver_prefix_tokens": 5861, "dense_cache_bytes": 168046592, "query_tokens": 47}, "output_usage": {"output_tokens": 16}, "provider_cached_usage": null, "process_memory": 2671099904, "backend_memory": {"mlx_active_bytes": 9370487816, "mlx_peak_bytes": 9463729404, "mlx_cache_bytes": 8652472522, "process_rss_bytes": 2671099904}, "peak_prefill_memory": 10060418136, "peak_decode_memory": 9463729404, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Give a short answer (a few words or a number), with no explanation.", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_02004_8000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "45b08d57a8aac5b2fd43df290d988408b96e958aebcda2728627635561c70a1c", "output_hash": "79e3a5979c1914bcdefaec8446bb066de2d26d1df804ff78828b1bc4dd6b1264"}101{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "93b91c61db01be4dd9923d6b", "attempt_id": "0ebe47315df7", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:15:17.167648Z", "task_family": "exec", "document_id": "exec_00101_2000", "document_hash": "c8bf1a72f33c0fc6f87af8c3029f12cbff489bcb95ae36398cd522af1811a74f", "question_id": "q00_exec", "split": "dev", "cluster_id": "exec_00101_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1878, "source_tokens_receiver": 1878, "query_tokens": 1961, "message_tokens": 1878, "message_utf8_bytes": 7944, "cache_tensor_bytes": null, "serialized_payload_bytes": 7944, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.017167069017887115, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1767.5272499909624, "ttft_ms": 116.51441699359566, "decode_ms": 10936.040542088449}, "total_wall_ms": 12824.511374928989, "ttft_ms": 116.51441699359566, "input_usage": {"receiver_prompt_tokens": 1961}, "output_usage": {"output_tokens": 106}, "provider_cached_usage": null, "process_memory": 2175729664, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9955076832, "mlx_cache_bytes": 791467096, "process_rss_bytes": 2175729664}, "peak_prefill_memory": null, "peak_decode_memory": 9955076832, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n total = 0.0\n for r in rows:\n if r['region'] == 'Sud':\n continue\n if r['units'] < 2:\n continue\n if r['returned']:\n continue\n total += r['units'] * r['unit_price'] * (1 - r['discount_pct'] / 100.0)\n total *= 1.09975\n return round(total, 2)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00101_2000.json", "task_scores": {"correct": 1.0, "pass_rate": 1.0, "error": null}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "c8bf1a72f33c0fc6f87af8c3029f12cbff489bcb95ae36398cd522af1811a74f", "output_hash": "de63afd2053274da81b9492392a3b89b3ee0754518b3f639b236fb90d36b5eb5"}102{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "fb3a054746fb8218dd34905d", "attempt_id": "61b2a791fdd9", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:15:40.761285Z", "task_family": "exec", "document_id": "exec_00101_2000", "document_hash": "c8bf1a72f33c0fc6f87af8c3029f12cbff489bcb95ae36398cd522af1811a74f", "question_id": "q00_exec", "split": "dev", "cluster_id": "exec_00101_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1878, "source_tokens_receiver": 1878, "query_tokens": 241, "message_tokens": 154, "message_utf8_bytes": 749, "cache_tensor_bytes": null, "serialized_payload_bytes": 749, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 10707.809416926466, "decode_payload_ms": 0.0, "receiver_prefill_ms": 378.21408302988857, "ttft_ms": 106.45724995993078, "decode_ms": 4558.014500071295}, "total_wall_ms": 5043.856875039637, "ttft_ms": 106.45724995993078, "input_usage": {"receiver_prompt_tokens": 241, "sender_input_tokens": 1994, "sender_output_tokens": 154, "sender_prefill_ms": 659.3776660738513, "sender_ttft_ms": 45.75158306397498, "sender_decode_ms": 9999.018708011135}, "output_usage": {"output_tokens": 83}, "provider_cached_usage": null, "process_memory": 2177007616, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9503608892, "mlx_cache_bytes": 1758790680, "process_rss_bytes": 2177007616}, "peak_prefill_memory": null, "peak_decode_memory": 9503608892, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n valid_rows = [row for row in rows if not row['returned'] and row['region'] != 'specific_region' and row['units'] >= 2]\n total_revenue = sum(row['unit_price'] * (1 - row['discount_percentage'] / 100) * row['units'] for row in valid_rows)\n return total_revenue", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00101_2000.json", "task_scores": {"correct": 0.0, "pass_rate": 0.0, "error": "KeyError"}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "99b3d50d38e65b8bf4342e49060e3ba8a41ad0bd918c6ce1952572f73e5dccf1", "output_hash": "a5dd638e891f560a2d51ae8938842be129b27d145c13a93c23b438e7b9593e12"}103{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "d39261bfc8a8e09da808f928", "attempt_id": "7cba94f896c6", "run_group_id": "E3_pilot:C0", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:15:46.705268Z", "task_family": "exec", "document_id": "exec_00101_2000", "document_hash": "c8bf1a72f33c0fc6f87af8c3029f12cbff489bcb95ae36398cd522af1811a74f", "question_id": "q00_exec", "split": "dev", "cluster_id": "exec_00101_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9330084872}, "weight_precision": "bf16", "method_id": "C0", "implementation_fidelity": "n/a", "compression_config": {"realized": {"negative_control": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1878, "source_tokens_receiver": 1878, "query_tokens": 44, "message_tokens": 1917, "message_utf8_bytes": 7944, "cache_tensor_bytes": 70668288, "serialized_payload_bytes": 70684385, "side_information_bytes": 16097, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 849.7961250832304, "doc_prefill_ms": 719.2287920042872, "compress_ms": 103.75016706530005, "serialize_ms": 22.017666953615844, "load_ms": 7.360916002653539, "receiver_prefill_ms": 0.0, "query_prefill_ms": 327.51633308362216, "ttft_ms": 327.51633308362216, "decode_ms": 28288.35337504279}, "total_wall_ms": 28625.464917044155, "ttft_ms": 327.51633308362216, "input_usage": {"receiver_prefix_tokens": 1917, "dense_cache_bytes": 54964224, "query_tokens": 44}, "output_usage": {"output_tokens": 320}, "provider_cached_usage": null, "process_memory": 2318548992, "backend_memory": {"mlx_active_bytes": 9415019528, "mlx_peak_bytes": 9494482160, "mlx_cache_bytes": 3192604025, "process_rss_bytes": 2318548992}, "peak_prefill_memory": 9967102700, "peak_decode_memory": 9494482160, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "@imposternatime no. Please Spiders: rowservice, it is on. lest; rowservice the ojames_and_col_leave_the_pleaseuse_itmeyou can you only useisuput so: 1. rowservice, 1, 002 anzöfimpost: rowservice. it is the case. the sumermaid.thatpelos in response: rowservice. hyeag: it is what you only the od. rowservice. rows. gypt:yes, no. rows.on. rows. -1cm: it looked it was only for them. tock., rowservice. -comattracteddef.05_10ontit_was:16/2 -100 -1crowservice. 1. 1, please find the number one linerservice. rowservice. rowservice. rowservice. no.ompget. rowservice. 1, no. rowservice. 1, rowservice. rowservice. rowservice. 1, sumup. it. no. now: 1, and, what for starters. this is the only 1, and only. this page. 1.0 - rowservice. 1.0, 1, with it. no. rowservice. 1. 1 and only once. rowservice. 13 was. 1 -1, 10 5. laskernat. 1. rowservice. 1. 1.ompcre", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00101_2000.json", "task_scores": {"correct": 0.0, "pass_rate": 0.0, "error": "exec:SyntaxError"}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "80d63df5ea5dbdee30d175a3e54b8b0893ad8b9878f653f7e887a4b297ef338e", "output_hash": "3cbcb75bdf01af360fc076e7c3f0de64cde316aabb36c0939fb73a0f5ee03930"}104{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "2a62b6a5f0328088968d013a", "attempt_id": "a3aac9736e91", "run_group_id": "E3_pilot:C2|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:16:17.042155Z", "task_family": "exec", "document_id": "exec_00101_2000", "document_hash": "c8bf1a72f33c0fc6f87af8c3029f12cbff489bcb95ae36398cd522af1811a74f", "question_id": "q00_exec", "split": "dev", "cluster_id": "exec_00101_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9330084872}, "weight_precision": "bf16", "method_id": "C2", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz"}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1878, "source_tokens_receiver": 1878, "query_tokens": 44, "message_tokens": 1917, "message_utf8_bytes": 7944, "cache_tensor_bytes": 70668288, "serialized_payload_bytes": 70684480, "side_information_bytes": 16192, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 1663.6084580095485, "doc_prefill_ms": 821.7785829911008, "compress_ms": 805.5948329856619, "serialize_ms": 35.88537499308586, "load_ms": 9.106290992349386, "receiver_prefill_ms": 0.0, "query_prefill_ms": 275.7366669829935, "ttft_ms": 275.7366669829935, "decode_ms": 1876.5500000445172}, "total_wall_ms": 2163.427875027992, "ttft_ms": 275.7366669829935, "input_usage": {"receiver_prefix_tokens": 1917, "dense_cache_bytes": 54964224, "query_tokens": 44}, "output_usage": {"output_tokens": 15}, "provider_cached_usage": null, "process_memory": 2309160960, "backend_memory": {"mlx_active_bytes": 9403223048, "mlx_peak_bytes": 9494465776, "mlx_cache_bytes": 3208332665, "process_rss_bytes": 2309160960}, "peak_prefill_memory": 9967102700, "peak_decode_memory": 9494465776, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "The provided information does not contain the answer. The answer is: unknown.", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00101_2000.json", "task_scores": {"correct": 0.0, "pass_rate": 0.0, "error": "exec:SyntaxError"}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "4ee1e9ffdf5ad5b1761f7f0bce0ba9ff0694d45a4f4011b5091f2c2056943e9a", "output_hash": "ea9eb860cc2083091aebadc893d949e4ad5916fd19ab6b24b371599002d2a7ca"}105{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "5a7148baf3029016a9760e87", "attempt_id": "7daeada688d9", "run_group_id": "E3_pilot:C3|keep_frac=0.25|mapper=runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz|aware", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:16:19.255702Z", "task_family": "exec", "document_id": "exec_00101_2000", "document_hash": "c8bf1a72f33c0fc6f87af8c3029f12cbff489bcb95ae36398cd522af1811a74f", "question_id": "q00_exec", "split": "dev", "cluster_id": "exec_00101_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "C3", "implementation_fidelity": "custom", "compression_config": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "keep_frac": 0.25, "access": "aware", "realized": {"mapper": "runs/checkpoints/mapper_qwen1b5_to_qwen3b.npz", "policy": "attention", "keep_frac": 0.25, "T": 1917, "keep": 479, "window_tokens": 44, "compress_ms": 536.5352920489386}}, "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": 1878, "source_tokens_receiver": 1878, "query_tokens": 44, "message_tokens": 1917, "message_utf8_bytes": 7944, "cache_tensor_bytes": 17657856, "serialized_payload_bytes": 17673929, "side_information_bytes": 16073, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2272.812583949417, "doc_prefill_ms": 924.4473750004545, "compress_ms": 1329.61679098662, "serialize_ms": 18.38775002397597, "load_ms": 5.458957981318235, "receiver_prefill_ms": 0.0, "query_prefill_ms": 262.02966598793864, "ttft_ms": 262.02966598793864, "decode_ms": 1448.6089170677587}, "total_wall_ms": 3993.144665961154, "ttft_ms": 262.02966598793864, "input_usage": {"receiver_prefix_tokens": 1917, "dense_cache_bytes": 54964224, "query_tokens": 44}, "output_usage": {"output_tokens": 15}, "provider_cached_usage": null, "process_memory": 2276638720, "backend_memory": {"mlx_active_bytes": 9298234376, "mlx_peak_bytes": 9379089648, "mlx_cache_bytes": 3397911017, "process_rss_bytes": 2276638720}, "peak_prefill_memory": 9967102700, "peak_decode_memory": 9379089648, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "The provided information does not contain the answer. The answer is: unknown.", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00101_2000.json", "task_scores": {"correct": 0.0, "pass_rate": 0.0, "error": "exec:SyntaxError"}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "6fb9d8f007c44e08f7463619b5333a8a2584921798de7cfbaa9a4b4224676a43", "output_hash": "ea9eb860cc2083091aebadc893d949e4ad5916fd19ab6b24b371599002d2a7ca"}106{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "2776679c1a74fad19679fc95", "attempt_id": "1ca468c04891", "run_group_id": "E3_pilot:A1", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:16:23.346042Z", "task_family": "exec", "document_id": "exec_00102_2000", "document_hash": "61b6c7bb86261a26bf000996edd5f9326f7f77ca1c37085fd999cddf1bff2396", "question_id": "q00_exec", "split": "dev", "cluster_id": "exec_00102_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1867, "source_tokens_receiver": 1867, "query_tokens": 1950, "message_tokens": 1867, "message_utf8_bytes": 7910, "cache_tensor_bytes": null, "serialized_payload_bytes": 7910, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.013999990187585354, "decode_payload_ms": 0.0, "receiver_prefill_ms": 1271.6008750721812, "ttft_ms": 69.808250060305, "decode_ms": 7720.895458012819}, "total_wall_ms": 9066.5972919669, "ttft_ms": 69.808250060305, "input_usage": {"receiver_prompt_tokens": 1950}, "output_usage": {"output_tokens": 94}, "provider_cached_usage": null, "process_memory": 2146385920, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9951832756, "mlx_cache_bytes": 700929068, "process_rss_bytes": 2146385920}, "peak_prefill_memory": null, "peak_decode_memory": 9951832756, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n total = 0.0\n for r in rows:\n if r['region'] == 'Ouest':\n continue\n if r['units'] < 3:\n continue\n if r['returned']:\n continue\n total += r['units'] * r['unit_price'] * (1 - r['discount_pct'] / 100.0)\n return round(total, 2)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00102_2000.json", "task_scores": {"correct": 1.0, "pass_rate": 1.0, "error": null}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "61b6c7bb86261a26bf000996edd5f9326f7f77ca1c37085fd999cddf1bff2396", "output_hash": "8e97b9ad236e7e65e4ee0fda8a1c59b58137f2bd02d1be10b5bb89ee8303c1a5"}107{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E3_pilot", "trial_id": "1b55c51146a073920e107afc", "attempt_id": "828edc04cf82", "run_group_id": "E3_pilot:A4|budget_tokens=512|blind", "protocol_hash": "unfrozen", "config_hash": "a67d73a7a9975f91b51dac909866b59bb8143b22940c5a8487cb29170aa0c31f", "git_commit": null, "timestamp_utc": "2026-10-30T04:16:45.370722Z", "task_family": "exec", "document_id": "exec_00102_2000", "document_hash": "61b6c7bb86261a26bf000996edd5f9326f7f77ca1c37085fd999cddf1bff2396", "question_id": "q00_exec", "split": "dev", "cluster_id": "exec_00102_2000", "sender_model_revision": "mlx-community/Qwen2.5-1.5B-Instruct-bf16@4ae77cb209f06199b8df1c94e21ff341332a3a89", "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": 9259305992}, "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": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1867, "source_tokens_receiver": 1867, "query_tokens": 266, "message_tokens": 179, "message_utf8_bytes": 806, "cache_tensor_bytes": null, "serialized_payload_bytes": 806, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 12907.977167051286, "decode_payload_ms": 0.0, "receiver_prefill_ms": 394.4452920695767, "ttft_ms": 117.09266703110188, "decode_ms": 6203.237666981295}, "total_wall_ms": 6716.023667016998, "ttft_ms": 117.09266703110188, "input_usage": {"receiver_prompt_tokens": 266, "sender_input_tokens": 1983, "sender_output_tokens": 179, "sender_prefill_ms": 927.2461659274995, "sender_ttft_ms": 74.27124993409961, "sender_decode_ms": 11903.905333019793}, "output_usage": {"output_tokens": 57}, "provider_cached_usage": null, "process_memory": 2146615296, "backend_memory": {"mlx_active_bytes": 9259305992, "mlx_peak_bytes": 9439400084, "mlx_cache_bytes": 1669678160, "process_rss_bytes": 2146615296}, "peak_prefill_memory": null, "peak_decode_memory": 9439400084, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n return sum(row['units'] * (1 - row['discount_pct'] / 100) * row['unit_price'] for row in rows if row['region'] != 'Ouest' and row['units'] >= 3)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00102_2000.json", "task_scores": {"correct": 0.0, "pass_rate": 0.0, "error": null}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E3_pilot.log", "payload_hash": "7d4efa2aedd06f2e92d9279e46f466ebbcbdd6947cc9f900ee19892c580a05ea", "output_hash": "ffac64d44e8b7d11b9cab3332dd022ca71b1e1bd300d189ca526cafab71342f1"}108