spb/wp30_uqo

Public
11commits
1branches
0releases
171.6 MBon disk
maindefault branch
3 days agolast push
203.8 KB · 65 lines
1{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "0cdeecde554f1bbc702ab2d7", "attempt_id": "940b84c62b36", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:23:56.618098Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 1978, "message_tokens": 1904, "message_utf8_bytes": 7128, "cache_tensor_bytes": null, "serialized_payload_bytes": 7128, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.008916947990655899, "decode_payload_ms": 0.0, "receiver_prefill_ms": 224.6155000757426, "ttft_ms": 14.26049997098744, "decode_ms": 147.60504104197025}, "total_wall_ms": 393.73458293266594, "ttft_ms": 14.26049997098744, "input_usage": {"receiver_prompt_tokens": 1978}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6598836224, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7035305764, "mlx_cache_bytes": 871836464, "process_rss_bytes": 6598836224}, "peak_prefill_memory": null, "peak_decode_memory": 7035305764, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "ZL-78K-5065", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "output_hash": "2a568d613468fe9e25fd3ab03aa2744ae45702a47a3665ea5aa63251d8bc749c"}2{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "16a351c32154fb80aafe4851", "attempt_id": "a77c99f6374f", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:23:57.030796Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 1983, "message_tokens": 1904, "message_utf8_bytes": 7128, "cache_tensor_bytes": null, "serialized_payload_bytes": 7128, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.008916947990655899, "decode_payload_ms": 0.0, "receiver_prefill_ms": 208.2602500449866, "ttft_ms": 13.479374931193888, "decode_ms": 120.93158403877169}, "total_wall_ms": 345.4456250183284, "ttft_ms": 13.479374931193888, "input_usage": {"receiver_prompt_tokens": 1983}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6598885376, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7037157440, "mlx_cache_bytes": 1651518256, "process_rss_bytes": 6598885376}, "peak_prefill_memory": null, "peak_decode_memory": 7037157440, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 19-N in Gaspé", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "output_hash": "42dd487ae6ba210216269d3f190a091ad80bb6727264e9b9b87bde124506854f"}3{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "5e69a27cae94ec18d3deeafd", "attempt_id": "47fb98cce6b4", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:23:57.392759Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 1982, "message_tokens": 1904, "message_utf8_bytes": 7128, "cache_tensor_bytes": null, "serialized_payload_bytes": 7128, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.008916947990655899, "decode_payload_ms": 0.0, "receiver_prefill_ms": 206.6167910816148, "ttft_ms": 13.585833949036896, "decode_ms": 93.64437498152256}, "total_wall_ms": 316.5831670630723, "ttft_ms": 13.585833949036896, "input_usage": {"receiver_prompt_tokens": 1982}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6598918144, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7036895296, "mlx_cache_bytes": 2000431920, "process_rss_bytes": 6598918144}, "peak_prefill_memory": null, "peak_decode_memory": 7036895296, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8485", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "output_hash": "d15ea89fae075fc945690894326d24c71a513cda40b6ede7884196520a0b2ad2"}4{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "4e2b583ab9a578e15093be33", "attempt_id": "f23795e304ce", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:04.787928Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 591, "message_tokens": 512, "message_utf8_bytes": 1087, "cache_tensor_bytes": null, "serialized_payload_bytes": 1087, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7061.945167020895, "decode_payload_ms": 0.0, "receiver_prefill_ms": 70.42479200754315, "ttft_ms": 14.768208027817309, "decode_ms": 13.10137496329844}, "total_wall_ms": 99.34099996462464, "ttft_ms": 14.768208027817309, "input_usage": {"receiver_prompt_tokens": 591, "sender_input_tokens": 2020, "sender_output_tokens": 512, "sender_prefill_ms": 211.29270899109542, "sender_ttft_ms": 13.70158395729959, "sender_decode_ms": 6835.405874997377}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599475200, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6564311128, "mlx_cache_bytes": 3381455664, "process_rss_bytes": 6599475200}, "peak_prefill_memory": null, "peak_decode_memory": 6564311128, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "762b58836b9acc90049d698b9be75bab93aa17c1904eb648b631e899f819781f", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}5{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "c25cf4513637ae8bb12bf782", "attempt_id": "6396a6bb1ce6", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:04.904204Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 596, "message_tokens": 512, "message_utf8_bytes": 1087, "cache_tensor_bytes": null, "serialized_payload_bytes": 1087, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7061.945167020895, "decode_payload_ms": 0.0, "receiver_prefill_ms": 69.93825000245124, "ttft_ms": 13.41466698795557, "decode_ms": 118.63916693255305}, "total_wall_ms": 202.99833302851766, "ttft_ms": 13.41466698795557, "input_usage": {"receiver_prompt_tokens": 596, "sender_input_tokens": 2020, "sender_output_tokens": 512, "sender_prefill_ms": 211.29270899109542, "sender_ttft_ms": 13.70158395729959, "sender_decode_ms": 6835.405874997377}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6599540736, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567358552, "mlx_cache_bytes": 3743804208, "process_rss_bytes": 6599540736}, "peak_prefill_memory": null, "peak_decode_memory": 6567358552, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 19-N, Gaspé", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "762b58836b9acc90049d698b9be75bab93aa17c1904eb648b631e899f819781f", "output_hash": "2e8cf782bd118a1c6d4e23417785d72fb92e930806b5752cd1ec405a0c9b95e5"}6{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "04f2e9121617b77e70933da5", "attempt_id": "428cff03217c", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:05.123683Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 595, "message_tokens": 512, "message_utf8_bytes": 1087, "cache_tensor_bytes": null, "serialized_payload_bytes": 1087, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7061.945167020895, "decode_payload_ms": 0.0, "receiver_prefill_ms": 69.41900006495416, "ttft_ms": 13.243000023066998, "decode_ms": 13.146916055120528}, "total_wall_ms": 96.77091601770371, "ttft_ms": 13.243000023066998, "input_usage": {"receiver_prompt_tokens": 595, "sender_input_tokens": 2020, "sender_output_tokens": 512, "sender_prefill_ms": 211.29270899109542, "sender_ttft_ms": 13.70158395729959, "sender_decode_ms": 6835.405874997377}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599557120, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6567161944, "mlx_cache_bytes": 3831393072, "process_rss_bytes": 6599557120}, "peak_prefill_memory": null, "peak_decode_memory": 6567161944, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "762b58836b9acc90049d698b9be75bab93aa17c1904eb648b631e899f819781f", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}7{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "2de07a8366b014f64a777955", "attempt_id": "54585b7311c2", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:12.270331Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 2, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 472, "message_tokens": 383, "message_utf8_bytes": 911, "cache_tensor_bytes": null, "serialized_payload_bytes": 911, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7033.391790930182, "decode_payload_ms": 0.0, "receiver_prefill_ms": 55.04641600418836, "ttft_ms": 13.070582994259894, "decode_ms": 13.27187498100102}, "total_wall_ms": 82.35450007487088, "ttft_ms": 13.070582994259894, "input_usage": {"receiver_prompt_tokens": 472, "sender_input_tokens": 2048, "sender_output_tokens": 512, "sender_prefill_ms": 212.1629579924047, "sender_ttft_ms": 13.434624997898936, "sender_decode_ms": 6806.032833061181}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599868416, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6483339188, "mlx_cache_bytes": 4948488844, "process_rss_bytes": 6599868416}, "peak_prefill_memory": null, "peak_decode_memory": 6483339188, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 2, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "a409b88ff601f1f9bf4d221a0c042117fadd2be19621e0b0dd2800495cfbe0cc", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}8{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "d86c0d30cb25c480c0dc0f8e", "attempt_id": "3cd3427514cb", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:12.369762Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 2, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 477, "message_tokens": 383, "message_utf8_bytes": 911, "cache_tensor_bytes": null, "serialized_payload_bytes": 911, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7033.391790930182, "decode_payload_ms": 0.0, "receiver_prefill_ms": 54.913791944272816, "ttft_ms": 13.033041032031178, "decode_ms": 143.7628340208903}, "total_wall_ms": 212.59883290622383, "ttft_ms": 13.033041032031178, "input_usage": {"receiver_prompt_tokens": 477, "sender_input_tokens": 2048, "sender_output_tokens": 512, "sender_prefill_ms": 212.1629579924047, "sender_ttft_ms": 13.434624997898936, "sender_decode_ms": 6806.032833061181}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6599983104, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6486304712, "mlx_cache_bytes": 5240142332, "process_rss_bytes": 6599983104}, "peak_prefill_memory": null, "peak_decode_memory": 6486304712, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 409-H in Senneterre", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 2, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.5}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "a409b88ff601f1f9bf4d221a0c042117fadd2be19621e0b0dd2800495cfbe0cc", "output_hash": "928c825f0fee3b69cf0901e3721b2d6c4467e52d2c6707afe17dd750a93d5e31"}9{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "e0fa6076b835b57d90e000e0", "attempt_id": "3341be8a605f", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:12.599065Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 2, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 476, "message_tokens": 383, "message_utf8_bytes": 911, "cache_tensor_bytes": null, "serialized_payload_bytes": 911, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7033.391790930182, "decode_payload_ms": 0.0, "receiver_prefill_ms": 54.901333060115576, "ttft_ms": 13.077417039312422, "decode_ms": 13.16474995110184}, "total_wall_ms": 82.01595896389335, "ttft_ms": 13.077417039312422, "input_usage": {"receiver_prompt_tokens": 476, "sender_input_tokens": 2048, "sender_output_tokens": 512, "sender_prefill_ms": 212.1629579924047, "sender_ttft_ms": 13.434624997898936, "sender_decode_ms": 6806.032833061181}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6599983104, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6486304712, "mlx_cache_bytes": 5240142332, "process_rss_bytes": 6599983104}, "peak_prefill_memory": null, "peak_decode_memory": 6486304712, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 2, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "a409b88ff601f1f9bf4d221a0c042117fadd2be19621e0b0dd2800495cfbe0cc", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}10{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "70d1869271e0b43fd4aa376f", "attempt_id": "09c299faeaa9", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:12.921678Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6319529992}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 35, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 71626752, "serialized_payload_bytes": 71642824, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 223.8553341012448, "doc_prefill_ms": 205.45991696417332, "compress_ms": 1.932834042236209, "serialize_ms": 13.479624991305172, "load_ms": 1.8750830786302686, "receiver_prefill_ms": 0.0, "query_prefill_ms": 26.82670799549669, "ttft_ms": 26.82670799549669, "decode_ms": 151.06991690117866}, "total_wall_ms": 181.1863750917837, "ttft_ms": 26.82670799549669, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 35}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6744637440, "backend_memory": {"mlx_active_bytes": 6393847816, "mlx_peak_bytes": 6438330572, "mlx_cache_bytes": 6120618773, "process_rss_bytes": 6744637440}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6438330572, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "ZL-78K-5065", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "50ad3bfd3b66fa37446a270ce342c602070ff789e5f2f92f155624b311440953", "output_hash": "2a568d613468fe9e25fd3ab03aa2744ae45702a47a3665ea5aa63251d8bc749c"}11{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "a29dc38143ae3cea948541d8", "attempt_id": "26715ee22645", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:13.119770Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6319529992}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 40, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 71626752, "serialized_payload_bytes": 71642824, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 223.8553341012448, "doc_prefill_ms": 205.45991696417332, "compress_ms": 1.932834042236209, "serialize_ms": 13.479624991305172, "load_ms": 1.7147089820355177, "receiver_prefill_ms": 0.0, "query_prefill_ms": 24.26987502258271, "ttft_ms": 24.26987502258271, "decode_ms": 122.9876249562949}, "total_wall_ms": 150.43770801275969, "ttft_ms": 24.26987502258271, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 40}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6744719360, "backend_memory": {"mlx_active_bytes": 6394044424, "mlx_peak_bytes": 6442262752, "mlx_cache_bytes": 6158662741, "process_rss_bytes": 6744719360}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6442262752, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 19-N in Gaspé", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "50ad3bfd3b66fa37446a270ce342c602070ff789e5f2f92f155624b311440953", "output_hash": "42dd487ae6ba210216269d3f190a091ad80bb6727264e9b9b87bde124506854f"}12{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "f98867b0a7ed6cf07067e982", "attempt_id": "6b3a3c19e902", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:13.287149Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6319529992}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 39, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 71626752, "serialized_payload_bytes": 71642824, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 223.8553341012448, "doc_prefill_ms": 205.45991696417332, "compress_ms": 1.932834042236209, "serialize_ms": 13.479624991305172, "load_ms": 1.7610830254852772, "receiver_prefill_ms": 0.0, "query_prefill_ms": 24.88725003786385, "ttft_ms": 24.88725003786385, "decode_ms": 95.83241608925164}, "total_wall_ms": 123.91741701867431, "ttft_ms": 24.88725003786385, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 39}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6744735744, "backend_memory": {"mlx_active_bytes": 6394044424, "mlx_peak_bytes": 6441967840, "mlx_cache_bytes": 6170524757, "process_rss_bytes": 6744735744}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6441967840, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8485", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "50ad3bfd3b66fa37446a270ce342c602070ff789e5f2f92f155624b311440953", "output_hash": "d15ea89fae075fc945690894326d24c71a513cda40b6ede7884196520a0b2ad2"}13{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "97b54fbb8a5290211601126e", "attempt_id": "e6a1ddee3c74", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:13.440758Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6319333384}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": "5fc169694cc6c011", "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 35, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 71626752, "serialized_payload_bytes": 71642852, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 12.695958954282105, "doc_prefill_ms": 205.45991696417332, "compress_ms": 0.6429170025512576, "serialize_ms": 11.892916983924806, "load_ms": 7.201208034530282, "receiver_prefill_ms": 0.0, "query_prefill_ms": 26.74016694072634, "ttft_ms": 26.74016694072634, "decode_ms": 150.1782500417903}, "total_wall_ms": 185.5770000256598, "ttft_ms": 26.74016694072634, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 35}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6816727040, "backend_memory": {"mlx_active_bytes": 6393847816, "mlx_peak_bytes": 6438166732, "mlx_cache_bytes": 6170721365, "process_rss_bytes": 6816727040}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6438166732, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "ZL-78K-5065", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "c51e89fbd3212a6c5beb924948522b88ce3815c4fabd2692b45795e28ab35bd1", "output_hash": "2a568d613468fe9e25fd3ab03aa2744ae45702a47a3665ea5aa63251d8bc749c"}14{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "f94f355d26e4252357688ff7", "attempt_id": "aa4dd37f1587", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:13.643029Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6319333384}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": "5fc169694cc6c011", "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 40, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 71626752, "serialized_payload_bytes": 71642852, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 12.695958954282105, "doc_prefill_ms": 205.45991696417332, "compress_ms": 0.6429170025512576, "serialize_ms": 11.892916983924806, "load_ms": 7.015291950665414, "receiver_prefill_ms": 0.0, "query_prefill_ms": 24.692957988008857, "ttft_ms": 24.692957988008857, "decode_ms": 122.84516601357609}, "total_wall_ms": 155.96595802344382, "ttft_ms": 24.692957988008857, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 40}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6822723584, "backend_memory": {"mlx_active_bytes": 6393847816, "mlx_peak_bytes": 6442098912, "mlx_cache_bytes": 6170721365, "process_rss_bytes": 6822723584}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6442098912, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 19-N in Gaspé", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "c51e89fbd3212a6c5beb924948522b88ce3815c4fabd2692b45795e28ab35bd1", "output_hash": "42dd487ae6ba210216269d3f190a091ad80bb6727264e9b9b87bde124506854f"}15{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "561870aba5f6d7fadabf2ccd", "attempt_id": "68858aaf69bc", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:13.815797Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6319333384}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": "5fc169694cc6c011", "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 39, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 71626752, "serialized_payload_bytes": 71642852, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 12.695958954282105, "doc_prefill_ms": 205.45991696417332, "compress_ms": 0.6429170025512576, "serialize_ms": 11.892916983924806, "load_ms": 6.962541025131941, "receiver_prefill_ms": 0.0, "query_prefill_ms": 25.750707951374352, "ttft_ms": 25.750707951374352, "decode_ms": 95.91704094782472}, "total_wall_ms": 130.02587505616248, "ttft_ms": 25.750707951374352, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 39}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6834716672, "backend_memory": {"mlx_active_bytes": 6393847816, "mlx_peak_bytes": 6441804000, "mlx_cache_bytes": 6170721365, "process_rss_bytes": 6834716672}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6441804000, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8485", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "c51e89fbd3212a6c5beb924948522b88ce3815c4fabd2692b45795e28ab35bd1", "output_hash": "d15ea89fae075fc945690894326d24c71a513cda40b6ede7884196520a0b2ad2"}16{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "1173cb4c5f3ba794471fdf39", "attempt_id": "2ebd39a7bf02", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:13.981902Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6286329736}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 35, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 38051712, "serialized_payload_bytes": 38085902, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 18.784375046379864, "doc_prefill_ms": 205.45991696417332, "compress_ms": 9.589166031219065, "serialize_ms": 8.846500073559582, "load_ms": 2.2877498995512724, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.24904196150601, "ttft_ms": 20.24904196150601, "decode_ms": 159.02462496887892}, "total_wall_ms": 183.02495800890028, "ttft_ms": 20.24904196150601, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 35}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6873759744, "backend_memory": {"mlx_active_bytes": 6326369800, "mlx_peak_bytes": 6382018856, "mlx_cache_bytes": 6415911717, "process_rss_bytes": 6873759744}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6382018856, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "ZL-78K-5065", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "6214db98b9866c4a46f9a487bc829c83429d2f6cd9158ee66ec01a6cd75fedba", "output_hash": "2a568d613468fe9e25fd3ab03aa2744ae45702a47a3665ea5aa63251d8bc749c"}17{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "c0730cd23c05700f031a228b", "attempt_id": "580a2255ad68", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:14.181951Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6286329736}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 40, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 38051712, "serialized_payload_bytes": 38085902, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 18.784375046379864, "doc_prefill_ms": 205.45991696417332, "compress_ms": 9.589166031219065, "serialize_ms": 8.846500073559582, "load_ms": 2.039124956354499, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.087040960788727, "ttft_ms": 20.087040960788727, "decode_ms": 129.69287496525794}, "total_wall_ms": 153.23804190848023, "ttft_ms": 20.087040960788727, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 40}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6874611712, "backend_memory": {"mlx_active_bytes": 6326373256, "mlx_peak_bytes": 6387713964, "mlx_cache_bytes": 6446797477, "process_rss_bytes": 6874611712}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6387713964, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 19-N in Gaspé", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "6214db98b9866c4a46f9a487bc829c83429d2f6cd9158ee66ec01a6cd75fedba", "output_hash": "42dd487ae6ba210216269d3f190a091ad80bb6727264e9b9b87bde124506854f"}18{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "0b11842279e677ce884166a0", "attempt_id": "e9b42e474ff7", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:14.352030Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6286329736}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 39, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 38051712, "serialized_payload_bytes": 38085902, "side_information_bytes": 34190, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 18.784375046379864, "doc_prefill_ms": 205.45991696417332, "compress_ms": 9.589166031219065, "serialize_ms": 8.846500073559582, "load_ms": 1.1808329727500677, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.33720805775374, "ttft_ms": 20.33720805775374, "decode_ms": 100.3128329757601}, "total_wall_ms": 123.31229099072516, "ttft_ms": 20.33720805775374, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 39}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6874677248, "backend_memory": {"mlx_active_bytes": 6326369800, "mlx_peak_bytes": 6386959916, "mlx_cache_bytes": 6469066789, "process_rss_bytes": 6874677248}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6386959916, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8485", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "6214db98b9866c4a46f9a487bc829c83429d2f6cd9158ee66ec01a6cd75fedba", "output_hash": "d15ea89fae075fc945690894326d24c71a513cda40b6ede7884196520a0b2ad2"}19{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "9af184f376b5983918a7645e", "attempt_id": "afda1c7d4c60", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:14.510152Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1943, "keep": 486, "compress_ms": 11.752249905839562}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 35, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 17915904, "serialized_payload_bytes": 17931816, "side_information_bytes": 15912, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 17.793874954804778, "doc_prefill_ms": 205.45991696417332, "compress_ms": 11.790457996539772, "serialize_ms": 5.850917077623308, "load_ms": 1.356916967779398, "receiver_prefill_ms": 0.0, "query_prefill_ms": 18.290749983862042, "ttft_ms": 18.290749983862042, "decode_ms": 144.2466670414433}, "total_wall_ms": 165.35562498029321, "ttft_ms": 18.290749983862042, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 35}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6891618304, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6322086092, "mlx_cache_bytes": 6553079125, "process_rss_bytes": 6891618304}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6322086092, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "ZL-78K-5065", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "2df7e1be733357d24b176adfd0157fc10cfca02f1b4ee6e7e51a6398620b9c86", "output_hash": "2a568d613468fe9e25fd3ab03aa2744ae45702a47a3665ea5aa63251d8bc749c"}20{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "b1c2f1615b679b4fab3e13e9", "attempt_id": "f6a2bcc28d24", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:14.692499Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1943, "keep": 486, "compress_ms": 11.752249905839562}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 40, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 17915904, "serialized_payload_bytes": 17931816, "side_information_bytes": 15912, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 17.793874954804778, "doc_prefill_ms": 205.45991696417332, "compress_ms": 11.790457996539772, "serialize_ms": 5.850917077623308, "load_ms": 1.4019169611856341, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.950583016499877, "ttft_ms": 17.950583016499877, "decode_ms": 52.652584039606154}, "total_wall_ms": 73.43124994076788, "ttft_ms": 17.950583016499877, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 40}, "output_usage": {"output_tokens": 4}, "provider_cached_usage": null, "process_memory": 6891618304, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6325526752, "mlx_cache_bytes": 6553079125, "process_rss_bytes": 6891618304}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6325526752, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Val-d'Or", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "2df7e1be733357d24b176adfd0157fc10cfca02f1b4ee6e7e51a6398620b9c86", "output_hash": "570448b634a06e2500dc0d88e08615836b31921efc964ef626fbe605c7de22f8"}21{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "155eb6fb7f777d591611a3f6", "attempt_id": "0c65afd6ff27", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:14.783520Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1943, "keep": 486, "compress_ms": 11.752249905839562}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1904, "source_tokens_receiver": 1904, "query_tokens": 39, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 17915904, "serialized_payload_bytes": 17931816, "side_information_bytes": 15912, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 17.793874954804778, "doc_prefill_ms": 205.45991696417332, "compress_ms": 11.790457996539772, "serialize_ms": 5.850917077623308, "load_ms": 1.4495420036837459, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.937000025995076, "ttft_ms": 17.937000025995076, "decode_ms": 92.08929201122373}, "total_wall_ms": 112.87679197266698, "ttft_ms": 17.937000025995076, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 39}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6891618304, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6325231840, "mlx_cache_bytes": 6553079125, "process_rss_bytes": 6891618304}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6325231840, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8485", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "2df7e1be733357d24b176adfd0157fc10cfca02f1b4ee6e7e51a6398620b9c86", "output_hash": "d15ea89fae075fc945690894326d24c71a513cda40b6ede7884196520a0b2ad2"}22{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "42e749db7c37b80a86266fe6", "attempt_id": "e18f8bcd6638", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:14.913133Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1943, "keep": 486, "window_tokens": 35, "compress_ms": 42.67583298496902}}, "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": 1904, "source_tokens_receiver": 1904, "query_tokens": 35, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 17915904, "serialized_payload_bytes": 17931855, "side_information_bytes": 15951, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 49.29812508635223, "doc_prefill_ms": 205.45991696417332, "compress_ms": 44.02137501165271, "serialize_ms": 5.06999995559454, "load_ms": 1.1016250355169177, "receiver_prefill_ms": 0.0, "query_prefill_ms": 18.07645906228572, "ttft_ms": 18.07645906228572, "decode_ms": 92.55154198035598}, "total_wall_ms": 162.5774159329012, "ttft_ms": 18.07645906228572, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 35}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6894206976, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6321594572, "mlx_cache_bytes": 6642208085, "process_rss_bytes": 6894206976}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6321594572, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "ZL-7806", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "ba6e95b7abdca0dfcd88e6219641a1e712510e789747f63c79172c956ae05f63", "output_hash": "7780cfcd22246240c403d82e763414b5a9fd57d08a2f73da0d4a3b1e9529e0df"}23{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "01656e0e5c4ba21451172c7e", "attempt_id": "cc4b453fe65b", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:15.092731Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1943, "keep": 486, "window_tokens": 40, "compress_ms": 41.40408290550113}}, "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": 1904, "source_tokens_receiver": 1904, "query_tokens": 40, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 17915904, "serialized_payload_bytes": 17931855, "side_information_bytes": 15951, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 47.68929199781269, "doc_prefill_ms": 205.45991696417332, "compress_ms": 42.87279106210917, "serialize_ms": 4.590249969623983, "load_ms": 1.2119589373469353, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.943665967322886, "ttft_ms": 17.943665967322886, "decode_ms": 117.44199995882809}, "total_wall_ms": 185.79029198735952, "ttft_ms": 17.943665967322886, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 40}, "output_usage": {"output_tokens": 9}, "provider_cached_usage": null, "process_memory": 6913753088, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6325526752, "mlx_cache_bytes": 6659804501, "process_rss_bytes": 6913753088}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6325526752, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 19-N in Gaspé", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "ba6e95b7abdca0dfcd88e6219641a1e712510e789747f63c79172c956ae05f63", "output_hash": "42dd487ae6ba210216269d3f190a091ad80bb6727264e9b9b87bde124506854f"}24{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "67c506a7c75f32c23caedf8c", "attempt_id": "3b8fd04ecc7a", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:15.295956Z", "task_family": "synthetic", "document_id": "syn_03000_2000", "document_hash": "c9cc27c20b05b1f6e592846e84990ee9b69c42f9720a8e338ec2f6fca777fdbc", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03000_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1943, "keep": 486, "window_tokens": 39, "compress_ms": 41.79983295034617}}, "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": 1904, "source_tokens_receiver": 1904, "query_tokens": 39, "message_tokens": 1943, "message_utf8_bytes": 7128, "cache_tensor_bytes": 17915904, "serialized_payload_bytes": 17931855, "side_information_bytes": 15951, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 47.19137505162507, "doc_prefill_ms": 205.45991696417332, "compress_ms": 43.17095794249326, "serialize_ms": 3.8068339927121997, "load_ms": 1.0715000098571181, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.849542084150016, "ttft_ms": 17.849542084150016, "decode_ms": 91.96720807813108}, "total_wall_ms": 159.61174992844462, "ttft_ms": 17.849542084150016, "input_usage": {"receiver_prefix_tokens": 1943, "dense_cache_bytes": 71626752, "query_tokens": 39}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6914359296, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6325231840, "mlx_cache_bytes": 6664670549, "process_rss_bytes": 6914359296}, "peak_prefill_memory": 7021657892, "peak_decode_memory": 6325231840, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "53.8585", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03000_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "ba6e95b7abdca0dfcd88e6219641a1e712510e789747f63c79172c956ae05f63", "output_hash": "c3c60563b91654a106b6ede0ec5f6914031e57a40b6674089be2b189adcdc1ab"}25{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "36443d5f00eb278b7f794230", "attempt_id": "8e837f512a91", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:15.508982Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 1941, "message_tokens": 1866, "message_utf8_bytes": 7342, "cache_tensor_bytes": null, "serialized_payload_bytes": 7342, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.00829098280519247, "decode_payload_ms": 0.0, "receiver_prefill_ms": 205.60774998739362, "ttft_ms": 13.509750016964972, "decode_ms": 133.14045895822346}, "total_wall_ms": 355.04633304663, "ttft_ms": 13.509750016964972, "input_usage": {"receiver_prompt_tokens": 1941}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6673530880, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7020134032, "mlx_cache_bytes": 856658952, "process_rss_bytes": 6673530880}, "peak_prefill_memory": null, "peak_decode_memory": 7020134032, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "GR-79S-5839", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "output_hash": "720cc40cd69e85e9728f6bea00b17a95ca07eee643e143a131840763958303cd"}26{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "df94c6ccbda4c7a58a545594", "attempt_id": "ae0e427ba429", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:15.880495Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 1945, "message_tokens": 1866, "message_utf8_bytes": 7342, "cache_tensor_bytes": null, "serialized_payload_bytes": 7342, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.00829098280519247, "decode_payload_ms": 0.0, "receiver_prefill_ms": 206.10354107338935, "ttft_ms": 13.530833995901048, "decode_ms": 146.94641600362957}, "total_wall_ms": 369.4087089970708, "ttft_ms": 13.530833995901048, "input_usage": {"receiver_prompt_tokens": 1945}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6673498112, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7021789248, "mlx_cache_bytes": 1621169160, "process_rss_bytes": 6673498112}, "peak_prefill_memory": null, "peak_decode_memory": 7021789248, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 344-B, Témiscaming", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "output_hash": "0197d491a5a04a142dc8e372360735a1d41757e7a5aea7e81f23b0e1ef6ee010"}27{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "71c4b2ecbe8464b625cf4286", "attempt_id": "10b3c7390a2a", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:16.266437Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 1945, "message_tokens": 1866, "message_utf8_bytes": 7342, "cache_tensor_bytes": null, "serialized_payload_bytes": 7342, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.00829098280519247, "decode_payload_ms": 0.0, "receiver_prefill_ms": 206.2315420480445, "ttft_ms": 13.486042036674917, "decode_ms": 93.44970900565386}, "total_wall_ms": 315.8948749769479, "ttft_ms": 13.486042036674917, "input_usage": {"receiver_prompt_tokens": 1945}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6673514496, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7021789248, "mlx_cache_bytes": 1621169160, "process_rss_bytes": 6673514496}, "peak_prefill_memory": null, "peak_decode_memory": 7021789248, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "55.7912", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "output_hash": "9aa9f6d48d43eb12185d973cf0d62919041cf70e6dee09af0dbc2dec0c8c1711"}28{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "e9eb7cdf54e34a509215a22a", "attempt_id": "167d5c1eeaf7", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:23.624845Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 591, "message_tokens": 512, "message_utf8_bytes": 1200, "cache_tensor_bytes": null, "serialized_payload_bytes": 1200, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7026.202415931039, "decode_payload_ms": 0.0, "receiver_prefill_ms": 70.36437501665205, "ttft_ms": 13.193292077630758, "decode_ms": 131.2020419863984}, "total_wall_ms": 215.8482091035694, "ttft_ms": 13.193292077630758, "input_usage": {"receiver_prompt_tokens": 591, "sender_input_tokens": 1982, "sender_output_tokens": 512, "sender_prefill_ms": 211.61324996501207, "sender_ttft_ms": 13.587541994638741, "sender_decode_ms": 6799.584458000027}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6673596416, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6564311128, "mlx_cache_bytes": 2986972168, "process_rss_bytes": 6673596416}, "peak_prefill_memory": null, "peak_decode_memory": 6564311128, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "GR-79S-5839", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "6d2a2bad9425191f4568f7eb115d63c953e27080a86953f62a982ad73340e770", "output_hash": "720cc40cd69e85e9728f6bea00b17a95ca07eee643e143a131840763958303cd"}29{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "719bf9696ebd729dd16e8677", "attempt_id": "eb14a9fe12ca", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:23.858277Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 595, "message_tokens": 512, "message_utf8_bytes": 1200, "cache_tensor_bytes": null, "serialized_payload_bytes": 1200, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7026.202415931039, "decode_payload_ms": 0.0, "receiver_prefill_ms": 70.23508299607784, "ttft_ms": 13.316791970282793, "decode_ms": 144.85295803751796}, "total_wall_ms": 229.4391670729965, "ttft_ms": 13.316791970282793, "input_usage": {"receiver_prompt_tokens": 595, "sender_input_tokens": 1982, "sender_output_tokens": 512, "sender_prefill_ms": 211.61324996501207, "sender_ttft_ms": 13.587541994638741, "sender_decode_ms": 6799.584458000027}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6673612800, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6566965336, "mlx_cache_bytes": 3348927496, "process_rss_bytes": 6673612800}, "peak_prefill_memory": null, "peak_decode_memory": 6566965336, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 344-B, Témiscaming", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "6d2a2bad9425191f4568f7eb115d63c953e27080a86953f62a982ad73340e770", "output_hash": "0197d491a5a04a142dc8e372360735a1d41757e7a5aea7e81f23b0e1ef6ee010"}30{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "51496ab7a3f13040feeeca32", "attempt_id": "8e7741f23b5b", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:24.104759Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 595, "message_tokens": 512, "message_utf8_bytes": 1200, "cache_tensor_bytes": null, "serialized_payload_bytes": 1200, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7026.202415931039, "decode_payload_ms": 0.0, "receiver_prefill_ms": 69.14679089095443, "ttft_ms": 13.034458039328456, "decode_ms": 13.002666993997991}, "total_wall_ms": 96.19120904244483, "ttft_ms": 13.034458039328456, "input_usage": {"receiver_prompt_tokens": 595, "sender_input_tokens": 1982, "sender_output_tokens": 512, "sender_prefill_ms": 211.61324996501207, "sender_ttft_ms": 13.587541994638741, "sender_decode_ms": 6799.584458000027}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6673629184, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6566965336, "mlx_cache_bytes": 3348927496, "process_rss_bytes": 6673629184}, "peak_prefill_memory": null, "peak_decode_memory": 6566965336, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "6d2a2bad9425191f4568f7eb115d63c953e27080a86953f62a982ad73340e770", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}31{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "3830178a5a91cd44d5dbf20d", "attempt_id": "4b1ff6f3134f", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:31.245103Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 443, "message_tokens": 352, "message_utf8_bytes": 830, "cache_tensor_bytes": null, "serialized_payload_bytes": 830, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7026.983082992956, "decode_payload_ms": 0.0, "receiver_prefill_ms": 54.0503750089556, "ttft_ms": 13.144208933226764, "decode_ms": 131.74858398269862}, "total_wall_ms": 199.8586249537766, "ttft_ms": 13.144208933226764, "input_usage": {"receiver_prompt_tokens": 443, "sender_input_tokens": 2010, "sender_output_tokens": 512, "sender_prefill_ms": 214.6057920763269, "sender_ttft_ms": 13.384083984419703, "sender_decode_ms": 6797.24666592665}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6673711104, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6542731096, "mlx_cache_bytes": 4537162480, "process_rss_bytes": 6673711104}, "peak_prefill_memory": null, "peak_decode_memory": 6542731096, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "GR-79S-5839", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "a613cf19d60b1dbe7d9f51bbbf98459d9f181f51f14c39d755912e891ae7fd2f", "output_hash": "720cc40cd69e85e9728f6bea00b17a95ca07eee643e143a131840763958303cd"}32{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "929d9212dcec2f4160b18c7f", "attempt_id": "2bb085e8a05c", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:31.462795Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 447, "message_tokens": 352, "message_utf8_bytes": 830, "cache_tensor_bytes": null, "serialized_payload_bytes": 830, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7026.983082992956, "decode_payload_ms": 0.0, "receiver_prefill_ms": 53.944542072713375, "ttft_ms": 13.022625003941357, "decode_ms": 12.960167019627988}, "total_wall_ms": 80.7943339459598, "ttft_ms": 13.022625003941357, "input_usage": {"receiver_prompt_tokens": 447, "sender_input_tokens": 2010, "sender_output_tokens": 512, "sender_prefill_ms": 214.6057920763269, "sender_ttft_ms": 13.384083984419703, "sender_decode_ms": 6797.24666592665}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6673727488, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6545925992, "mlx_cache_bytes": 4886733288, "process_rss_bytes": 6673727488}, "peak_prefill_memory": null, "peak_decode_memory": 6545925992, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "a613cf19d60b1dbe7d9f51bbbf98459d9f181f51f14c39d755912e891ae7fd2f", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}33{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "143ef1aae63741be7d6e3398", "attempt_id": "69f99fcdf9d9", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:31.560797Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 8, "unresolved_quotes": 0, "parse_ok": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 447, "message_tokens": 352, "message_utf8_bytes": 830, "cache_tensor_bytes": null, "serialized_payload_bytes": 830, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7026.983082992956, "decode_payload_ms": 0.0, "receiver_prefill_ms": 54.1125419549644, "ttft_ms": 13.09975003823638, "decode_ms": 13.059709104709327}, "total_wall_ms": 81.10374992247671, "ttft_ms": 13.09975003823638, "input_usage": {"receiver_prompt_tokens": 447, "sender_input_tokens": 2010, "sender_output_tokens": 512, "sender_prefill_ms": 214.6057920763269, "sender_ttft_ms": 13.384083984419703, "sender_decode_ms": 6797.24666592665}, "output_usage": {"output_tokens": 1}, "provider_cached_usage": null, "process_memory": 6673727488, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6545925992, "mlx_cache_bytes": 4886733288, "process_rss_bytes": 6673727488}, "peak_prefill_memory": null, "peak_decode_memory": 6545925992, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "unknown", "evidence_artifact": {"n_facts": 8, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "a613cf19d60b1dbe7d9f51bbbf98459d9f181f51f14c39d755912e891ae7fd2f", "output_hash": "b23a6a8439c0dde5515893e7c90c1e3233b8616e634470f20dc4928bcf3609bc"}34{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "bb70bb2eff4f381b69ee42e8", "attempt_id": "2f06bbce413c", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:31.882176Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6318350344}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 36, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 70225920, "serialized_payload_bytes": 70241992, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 223.4000840689987, "doc_prefill_ms": 206.30833297036588, "compress_ms": 1.671832986176014, "serialize_ms": 12.567208032123744, "load_ms": 1.7305410001426935, "receiver_prefill_ms": 0.0, "query_prefill_ms": 25.622625020332634, "ttft_ms": 25.622625020332634, "decode_ms": 136.90641697030514}, "total_wall_ms": 165.638874983415, "ttft_ms": 25.622625020332634, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 36}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6814793728, "backend_memory": {"mlx_active_bytes": 6390472712, "mlx_peak_bytes": 6436511952, "mlx_cache_bytes": 5649244937, "process_rss_bytes": 6814793728}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6436511952, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "GR-79S-5839", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "3677b279bf1930a84687ecf30c0b771e10996ab18cad71e7eae40f759dc14f1f", "output_hash": "720cc40cd69e85e9728f6bea00b17a95ca07eee643e143a131840763958303cd"}35{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "f47352133653652ee7291581", "attempt_id": "9c4b7eeb80a4", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:32.064844Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6318350344}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 70225920, "serialized_payload_bytes": 70241992, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 223.4000840689987, "doc_prefill_ms": 206.30833297036588, "compress_ms": 1.671832986176014, "serialize_ms": 12.567208032123744, "load_ms": 1.9025420770049095, "receiver_prefill_ms": 0.0, "query_prefill_ms": 25.905083981342614, "ttft_ms": 25.905083981342614, "decode_ms": 149.9694170197472}, "total_wall_ms": 179.2355839861557, "ttft_ms": 25.905083981342614, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6814793728, "backend_memory": {"mlx_active_bytes": 6391586824, "mlx_peak_bytes": 6439674080, "mlx_cache_bytes": 5759509577, "process_rss_bytes": 6814793728}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6439674080, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 344-B, Témiscaming", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "3677b279bf1930a84687ecf30c0b771e10996ab18cad71e7eae40f759dc14f1f", "output_hash": "0197d491a5a04a142dc8e372360735a1d41757e7a5aea7e81f23b0e1ef6ee010"}36{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "fb133b8be1713e4c16308356", "attempt_id": "ae4e4d09ed29", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:32.261226Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6318350344}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 70225920, "serialized_payload_bytes": 70241992, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 223.4000840689987, "doc_prefill_ms": 206.30833297036588, "compress_ms": 1.671832986176014, "serialize_ms": 12.567208032123744, "load_ms": 1.4766668900847435, "receiver_prefill_ms": 0.0, "query_prefill_ms": 25.805040961131454, "ttft_ms": 25.805040961131454, "decode_ms": 95.94579203985631}, "total_wall_ms": 124.58670791238546, "ttft_ms": 25.805040961131454, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6814793728, "backend_memory": {"mlx_active_bytes": 6390505480, "mlx_peak_bytes": 6439674080, "mlx_cache_bytes": 5760590921, "process_rss_bytes": 6814793728}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6439674080, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "55.7912", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "3677b279bf1930a84687ecf30c0b771e10996ab18cad71e7eae40f759dc14f1f", "output_hash": "9aa9f6d48d43eb12185d973cf0d62919041cf70e6dee09af0dbc2dec0c8c1711"}37{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "42a8fab6e20cd8734e717c8e", "attempt_id": "4a197cc336eb", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:32.415048Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6318153736}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": "50f7e52278544a92", "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 36, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 70225920, "serialized_payload_bytes": 70242020, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 12.288458994589746, "doc_prefill_ms": 206.30833297036588, "compress_ms": 0.7226249435916543, "serialize_ms": 11.415082961320877, "load_ms": 7.172917015850544, "receiver_prefill_ms": 0.0, "query_prefill_ms": 25.97345900721848, "ttft_ms": 25.97345900721848, "decode_ms": 136.07970799785107}, "total_wall_ms": 170.71112501434982, "ttft_ms": 25.97345900721848, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 36}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6885572608, "backend_memory": {"mlx_active_bytes": 6390112264, "mlx_peak_bytes": 6436151504, "mlx_cache_bytes": 5760984137, "process_rss_bytes": 6885572608}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6436151504, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "GR-79S-5839", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "4d8c7fb29b1e81e6aa157b9c3a76ef18b3649b33ce29e5382a51b8137bcb9a24", "output_hash": "720cc40cd69e85e9728f6bea00b17a95ca07eee643e143a131840763958303cd"}38{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "e2816d81cf1176419c4d79ac", "attempt_id": "9db090a9881e", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:32.602788Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6318153736}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": "50f7e52278544a92", "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 70225920, "serialized_payload_bytes": 70242020, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 12.288458994589746, "doc_prefill_ms": 206.30833297036588, "compress_ms": 0.7226249435916543, "serialize_ms": 11.415082961320877, "load_ms": 6.364792003296316, "receiver_prefill_ms": 0.0, "query_prefill_ms": 26.193207944743335, "ttft_ms": 26.193207944743335, "decode_ms": 150.40091704577208}, "total_wall_ms": 184.37645805533975, "ttft_ms": 26.193207944743335, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6891470848, "backend_memory": {"mlx_active_bytes": 6391390216, "mlx_peak_bytes": 6439510240, "mlx_cache_bytes": 5759706185, "process_rss_bytes": 6891470848}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6439510240, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 344-B, Témiscaming", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "4d8c7fb29b1e81e6aa157b9c3a76ef18b3649b33ce29e5382a51b8137bcb9a24", "output_hash": "0197d491a5a04a142dc8e372360735a1d41757e7a5aea7e81f23b0e1ef6ee010"}39{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "9da6152e9427a2caffe6ab13", "attempt_id": "9e5e138905c7", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:32.804483Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6318153736}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": "50f7e52278544a92", "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 70225920, "serialized_payload_bytes": 70242020, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 12.288458994589746, "doc_prefill_ms": 206.30833297036588, "compress_ms": 0.7226249435916543, "serialize_ms": 11.415082961320877, "load_ms": 6.330708973109722, "receiver_prefill_ms": 0.0, "query_prefill_ms": 26.160874986089766, "ttft_ms": 26.160874986089766, "decode_ms": 96.01337497588247}, "total_wall_ms": 129.90116700530052, "ttft_ms": 26.160874986089766, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6903267328, "backend_memory": {"mlx_active_bytes": 6390112264, "mlx_peak_bytes": 6439510240, "mlx_cache_bytes": 5760984137, "process_rss_bytes": 6903267328}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6439510240, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "55.7912", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "4d8c7fb29b1e81e6aa157b9c3a76ef18b3649b33ce29e5382a51b8137bcb9a24", "output_hash": "9aa9f6d48d43eb12185d973cf0d62919041cf70e6dee09af0dbc2dec0c8c1711"}40{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "6e7db4b8ccd9f063dd6f1380", "attempt_id": "722f9f82746a", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:32.971916Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6284958856}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 36, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 37307520, "serialized_payload_bytes": 37341702, "side_information_bytes": 34182, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20.73895907960832, "doc_prefill_ms": 206.30833297036588, "compress_ms": 9.690957958810031, "serialize_ms": 10.710833012126386, "load_ms": 1.9134159665554762, "receiver_prefill_ms": 0.0, "query_prefill_ms": 19.98700003605336, "ttft_ms": 19.98700003605336, "decode_ms": 144.2867909790948}, "total_wall_ms": 167.8441249532625, "ttft_ms": 19.98700003605336, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 36}, "output_usage": {"output_tokens": 10}, "provider_cached_usage": null, "process_memory": 6940262400, "backend_memory": {"mlx_active_bytes": 6323922952, "mlx_peak_bytes": 6380754996, "mlx_cache_bytes": 5989221801, "process_rss_bytes": 6940262400}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6380754996, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "GR-79S-5839", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "4dfc3b91fab3c041bcae1d5c34775b35134e4b04a73949fb7b5b71be7ab3cc7e", "output_hash": "720cc40cd69e85e9728f6bea00b17a95ca07eee643e143a131840763958303cd"}41{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "e671799fdfd860e62add8a9b", "attempt_id": "70ab106fad3f", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:33.156453Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6284958856}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 37307520, "serialized_payload_bytes": 37341702, "side_information_bytes": 34182, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20.73895907960832, "doc_prefill_ms": 206.30833297036588, "compress_ms": 9.690957958810031, "serialize_ms": 10.710833012126386, "load_ms": 1.5715829795226455, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.322666969150305, "ttft_ms": 20.322666969150305, "decode_ms": 158.29691709950566}, "total_wall_ms": 181.60620890557766, "ttft_ms": 20.322666969150305, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6940311552, "backend_memory": {"mlx_active_bytes": 6323928712, "mlx_peak_bytes": 6385713100, "mlx_cache_bytes": 6024037033, "process_rss_bytes": 6940311552}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6385713100, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 344-B, Témiscaming", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "4dfc3b91fab3c041bcae1d5c34775b35134e4b04a73949fb7b5b71be7ab3cc7e", "output_hash": "0197d491a5a04a142dc8e372360735a1d41757e7a5aea7e81f23b0e1ef6ee010"}42{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "c66cd7915425eaf9bf5ef305", "attempt_id": "9f1d168fd516", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:33.355542Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6284958856}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 37307520, "serialized_payload_bytes": 37341702, "side_information_bytes": 34182, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 20.73895907960832, "doc_prefill_ms": 206.30833297036588, "compress_ms": 9.690957958810031, "serialize_ms": 10.710833012126386, "load_ms": 2.5863749906420708, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.015374990180135, "ttft_ms": 20.015374990180135, "decode_ms": 100.65812501125038}, "total_wall_ms": 124.7508330270648, "ttft_ms": 20.015374990180135, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6940327936, "backend_memory": {"mlx_active_bytes": 6323924104, "mlx_peak_bytes": 6385713868, "mlx_cache_bytes": 6024041641, "process_rss_bytes": 6940327936}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6385713868, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "55.7912", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "4dfc3b91fab3c041bcae1d5c34775b35134e4b04a73949fb7b5b71be7ab3cc7e", "output_hash": "9aa9f6d48d43eb12185d973cf0d62919041cf70e6dee09af0dbc2dec0c8c1711"}43{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "f0419355a2b518cd98b7508d", "attempt_id": "cacbf0063062", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:33.509638Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1905, "keep": 476, "compress_ms": 6.862125010229647}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 36, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 17547264, "serialized_payload_bytes": 17563173, "side_information_bytes": 15909, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 11.645708000287414, "doc_prefill_ms": 206.30833297036588, "compress_ms": 6.897040992043912, "serialize_ms": 4.575916100293398, "load_ms": 1.2702500680461526, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.405332997441292, "ttft_ms": 17.405332997441292, "decode_ms": 145.7374170422554}, "total_wall_ms": 165.90966598596424, "ttft_ms": 17.405332997441292, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 36}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6957793280, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6320791760, "mlx_cache_bytes": 6123302921, "process_rss_bytes": 6957793280}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6320791760, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "UW-43X-7145", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "2c3ce03cd14d33918caedba306c910277ca3abbc3aee9acb880677404489d769", "output_hash": "15e44e464b359da08c5330183a95e89f2f833f17a6b23b258cf155b193215141"}44{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "702ed819104155f4796f66d3", "attempt_id": "30590278389f", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:33.692510Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1905, "keep": 476, "compress_ms": 6.862125010229647}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 17547264, "serialized_payload_bytes": 17563173, "side_information_bytes": 15909, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 11.645708000287414, "doc_prefill_ms": 206.30833297036588, "compress_ms": 6.897040992043912, "serialize_ms": 4.575916100293398, "load_ms": 0.8646249771118164, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.75741600431502, "ttft_ms": 17.75741600431502, "decode_ms": 26.31945803295821}, "total_wall_ms": 46.301124966703355, "ttft_ms": 17.75741600431502, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 2}, "provider_cached_usage": null, "process_memory": 6957793280, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6325297376, "mlx_cache_bytes": 6123302921, "process_rss_bytes": 6957793280}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6325297376, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Not provided", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 0.0, "abstained": 1.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "2c3ce03cd14d33918caedba306c910277ca3abbc3aee9acb880677404489d769", "output_hash": "d83b9ff3c07a8390e2ea0a2af45fba37595da25f31b85ba625363ee98c0c796f"}45{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "3df5b7a9d4c63f92780cc0b2", "attempt_id": "396e6e1d8c15", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:33.755720Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1905, "keep": 476, "compress_ms": 6.862125010229647}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 3, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 17547264, "serialized_payload_bytes": 17563173, "side_information_bytes": 15909, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 11.645708000287414, "doc_prefill_ms": 206.30833297036588, "compress_ms": 6.897040992043912, "serialize_ms": 4.575916100293398, "load_ms": 1.4882499817758799, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.87608303129673, "ttft_ms": 17.87608303129673, "decode_ms": 91.80658403784037}, "total_wall_ms": 112.56412498187274, "ttft_ms": 17.87608303129673, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6957793280, "backend_memory": {"mlx_active_bytes": 6286303240, "mlx_peak_bytes": 6325297376, "mlx_cache_bytes": 6123302921, "process_rss_bytes": 6957793280}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6325297376, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "55.7912", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "2c3ce03cd14d33918caedba306c910277ca3abbc3aee9acb880677404489d769", "output_hash": "9aa9f6d48d43eb12185d973cf0d62919041cf70e6dee09af0dbc2dec0c8c1711"}46{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "47cf3cfa027b2873c417e673", "attempt_id": "dcf628c7d1da", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:33.885867Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q00_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1905, "keep": 476, "window_tokens": 36, "compress_ms": 40.79966701101512}}, "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": 1866, "source_tokens_receiver": 1866, "query_tokens": 36, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 17547264, "serialized_payload_bytes": 17563213, "side_information_bytes": 15949, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 46.04279191698879, "doc_prefill_ms": 206.30833297036588, "compress_ms": 42.15516592375934, "serialize_ms": 3.6762079689651728, "load_ms": 0.4534169565886259, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.04520802013576, "ttft_ms": 17.04520802013576, "decode_ms": 91.69720800127834}, "total_wall_ms": 156.79966600146145, "ttft_ms": 17.04520802013576, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 36}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6958825472, "backend_memory": {"mlx_active_bytes": 6285123592, "mlx_peak_bytes": 6319808720, "mlx_cache_bytes": 6212006033, "process_rss_bytes": 6958825472}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6319808720, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "GR-7993-S", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 0.0, "abstained": 0.0, "f1": 0.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "de01306daf2ef2463b26fb5b78ffc7f9d192dc8025fa4074f1108bebf5ad6ffd", "output_hash": "ae7c1ebd5fa193e488ec03fceef4084c4647ba987220f71f000e452cc464314a"}47{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "d4121b7b1a6d290fe9e93519", "attempt_id": "e5fd2823ac2c", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:34.060122Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q01_single_fact", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1905, "keep": 476, "window_tokens": 40, "compress_ms": 40.86537496186793}}, "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": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 17547264, "serialized_payload_bytes": 17563213, "side_information_bytes": 15949, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 46.05379200074822, "doc_prefill_ms": 206.30833297036588, "compress_ms": 42.205333011224866, "serialize_ms": 3.642000025138259, "load_ms": 1.1582919396460056, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.554208054207265, "ttft_ms": 17.554208054207265, "decode_ms": 144.71179200336337}, "total_wall_ms": 211.04904101230204, "ttft_ms": 17.554208054207265, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 11}, "provider_cached_usage": null, "process_memory": 6959464448, "backend_memory": {"mlx_active_bytes": 6285123592, "mlx_peak_bytes": 6324347104, "mlx_cache_bytes": 6229504145, "process_rss_bytes": 6959464448}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6324347104, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "Lot 344-B, Témiscaming", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 0.8571428571428571}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "de01306daf2ef2463b26fb5b78ffc7f9d192dc8025fa4074f1108bebf5ad6ffd", "output_hash": "0197d491a5a04a142dc8e372360735a1d41757e7a5aea7e81f23b0e1ef6ee010"}48{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "398ba8de7f60a175617318d3", "attempt_id": "a6cb612f8995", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:34.288940Z", "task_family": "synthetic", "document_id": "syn_03003_2000", "document_hash": "dddca1a38e3224870fcfae927c701c400fd885c19123ea840ee06a76a140f407", "question_id": "q02_exact_copy", "split": "test", "cluster_id": "syn_03003_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1905, "keep": 476, "window_tokens": 40, "compress_ms": 39.93862494826317}}, "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": 1866, "source_tokens_receiver": 1866, "query_tokens": 40, "message_tokens": 1905, "message_utf8_bytes": 7342, "cache_tensor_bytes": 17547264, "serialized_payload_bytes": 17563213, "side_information_bytes": 15949, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 44.9476670473814, "doc_prefill_ms": 206.30833297036588, "compress_ms": 41.25804197974503, "serialize_ms": 3.4895000280812383, "load_ms": 1.231792033649981, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.622416955418885, "ttft_ms": 17.622416955418885, "decode_ms": 91.71945904381573}, "total_wall_ms": 157.11733407806605, "ttft_ms": 17.622416955418885, "input_usage": {"receiver_prefix_tokens": 1905, "dense_cache_bytes": 70225920, "query_tokens": 40}, "output_usage": {"output_tokens": 7}, "provider_cached_usage": null, "process_memory": 6959988736, "backend_memory": {"mlx_active_bytes": 6285123592, "mlx_peak_bytes": 6324347104, "mlx_cache_bytes": 6229504145, "process_rss_bytes": 6959988736}, "peak_prefill_memory": 7006764688, "peak_decode_memory": 6324347104, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "55.7912", "evidence_artifact": null, "reference_key_id": "keys/synthetic/syn_03003_2000.json", "task_scores": {"correct": 1.0, "abstained": 0.0, "f1": 1.0}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "de01306daf2ef2463b26fb5b78ffc7f9d192dc8025fa4074f1108bebf5ad6ffd", "output_hash": "9aa9f6d48d43eb12185d973cf0d62919041cf70e6dee09af0dbc2dec0c8c1711"}49{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "2658de767d328e91a4056217", "attempt_id": "54e85575c0d9", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:34.498693Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 1926, "message_tokens": 1843, "message_utf8_bytes": 7892, "cache_tensor_bytes": null, "serialized_payload_bytes": 7892, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.008500064723193645, "decode_payload_ms": 0.0, "receiver_prefill_ms": 208.54691695421934, "ttft_ms": 13.47845804411918, "decode_ms": 1395.0450830161572}, "total_wall_ms": 1619.954333989881, "ttft_ms": 13.47845804411918, "input_usage": {"receiver_prompt_tokens": 1926}, "output_usage": {"output_tokens": 105}, "provider_cached_usage": null, "process_memory": 6743097344, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7014645332, "mlx_cache_bytes": 851170252, "process_rss_bytes": 6743097344}, "peak_prefill_memory": null, "peak_decode_memory": 7014645332, "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['returned']:\n            continue\n        if r['region'] == 'Centre':\n            continue\n        if r['units'] < 4:\n            continue\n        total += r['units'] * r['unit_price'] * (1 - r['discount_pct'] / 100.0)\n    total *= 1.09975\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "output_hash": "5f25173e7b1a616c4eea703e50311df5ebdd9ac271873ce14a3700496d807d0c"}50{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "1c8aa5cb02a8b9f6f8bb2be3", "attempt_id": "7f063b9ef737", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:38.922812Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "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": 1843, "source_tokens_receiver": 1843, "query_tokens": 278, "message_tokens": 191, "message_utf8_bytes": 780, "cache_tensor_bytes": null, "serialized_payload_bytes": 780, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 2767.087584012188, "decode_payload_ms": 0.0, "receiver_prefill_ms": 42.84758400171995, "ttft_ms": 13.182707945816219, "decode_ms": 1562.9609590396285}, "total_wall_ms": 1619.9037079932168, "ttft_ms": 13.182707945816219, "input_usage": {"receiver_prompt_tokens": 278, "sender_input_tokens": 1959, "sender_output_tokens": 191, "sender_prefill_ms": 211.38341608457267, "sender_ttft_ms": 13.397792004980147, "sender_decode_ms": 2540.150915971026}, "output_usage": {"output_tokens": 120}, "provider_cached_usage": null, "process_memory": 6743130112, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6411625668, "mlx_cache_bytes": 1970198560, "process_rss_bytes": 6743130112}, "peak_prefill_memory": null, "peak_decode_memory": 6411625668, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n    total_revenue = 0\n    tax_rate = 0.09975\n    for row in rows:\n        if row['region'] != 'Centre' and row['units'] >= 4 and not row['returned']:\n            line_revenue = row['units'] * row['unit_price'] * (1 - row['discount_pct'] / 100)\n            total_revenue += round(line_revenue, 0)\n    net_revenue = total_revenue * (1 + tax_rate)\n    return net_revenue", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "06b3b81f0a47ff3a2a9c1e4595121c73dada56c05a6004234ce14ba5764dfb6f", "output_hash": "afbb72ea32fd836155366577993b67fc39cd67a518204719cc03fb45dfdb8fcc"}51{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "efd06228498ba4c93d414e96", "attempt_id": "9b540bd74966", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:46.217636Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 9, "unresolved_quotes": 0, "parse_ok": 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": 1843, "source_tokens_receiver": 1843, "query_tokens": 474, "message_tokens": 375, "message_utf8_bytes": 1068, "cache_tensor_bytes": null, "serialized_payload_bytes": 1068, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5637.319291941822, "decode_payload_ms": 0.0, "receiver_prefill_ms": 55.35399995278567, "ttft_ms": 13.052541995421052, "decode_ms": 771.0082499543205}, "total_wall_ms": 840.440833941102, "ttft_ms": 13.052541995421052, "input_usage": {"receiver_prompt_tokens": 474, "sender_input_tokens": 1987, "sender_output_tokens": 407, "sender_prefill_ms": 213.2203329820186, "sender_ttft_ms": 13.556707999669015, "sender_decode_ms": 5408.504665945657}, "output_usage": {"output_tokens": 59}, "provider_cached_usage": null, "process_memory": 6743162880, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6485485500, "mlx_cache_bytes": 3169820548, "process_rss_bytes": 6743162880}, "peak_prefill_memory": null, "peak_decode_memory": 6485485500, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n    return [row['revenue'] * (1 - 0.09975) for row in rows if row['units'] >= 4 and row['severity'] != 2 and row['order_date'] < row['shipment_date']]", "evidence_artifact": {"n_facts": 9, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "0be4160b7d4041cf0ea9d923002439a45ccedcdbff92c1fc1f54d93f74fceccd", "output_hash": "a36650a44ab48b595ce017e3c32d3db4c25ca59ea9f9c024edf857850898502a"}52{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "0a5ddb1d5770f38625f20406", "attempt_id": "bba41a0f4ca5", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:47.315183Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6316974088}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7892, "cache_tensor_bytes": 69378048, "serialized_payload_bytes": 69394120, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 219.41220795270056, "doc_prefill_ms": 202.10316707380116, "compress_ms": 1.669999910518527, "serialize_ms": 12.34512496739626, "load_ms": 1.9585000118240714, "receiver_prefill_ms": 0.0, "query_prefill_ms": 26.22825000435114, "ttft_ms": 26.22825000435114, "decode_ms": 1431.854874943383}, "total_wall_ms": 1461.8850419064984, "ttft_ms": 26.22825000435114, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 105}, "provider_cached_usage": null, "process_memory": 6811926528, "backend_memory": {"mlx_active_bytes": 6392471560, "mlx_peak_bytes": 6441935088, "mlx_cache_bytes": 4179142773, "process_rss_bytes": 6811926528}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6441935088, "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['returned']:\n            continue\n        if r['region'] == 'Centre':\n            continue\n        if r['units'] < 4:\n            continue\n        total += r['units'] * r['unit_price'] * (1 - r['discount_pct'] / 100.0)\n    total *= 1.09975\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "3036b65d7b530e49ba6e53370a2704f0f4894b07606a6893703eda3bfbd60971", "output_hash": "5f25173e7b1a616c4eea703e50311df5ebdd9ac271873ce14a3700496d807d0c"}53{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "d04058fe7c1089e97bf2790d", "attempt_id": "01497d79f99f", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:48.827497Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6316974088}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": "8f144c01b45c8d14", "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7892, "cache_tensor_bytes": 69378048, "serialized_payload_bytes": 69394148, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 13.70500004850328, "doc_prefill_ms": 202.10316707380116, "compress_ms": 1.5941669698804617, "serialize_ms": 11.92249997984618, "load_ms": 6.844499963335693, "receiver_prefill_ms": 0.0, "query_prefill_ms": 26.606291998177767, "ttft_ms": 26.606291998177767, "decode_ms": 1440.4833340086043}, "total_wall_ms": 1475.3914170432836, "ttft_ms": 26.606291998177767, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 105}, "provider_cached_usage": null, "process_memory": 6881525760, "backend_memory": {"mlx_active_bytes": 6392471560, "mlx_peak_bytes": 6441738480, "mlx_cache_bytes": 4179142773, "process_rss_bytes": 6881525760}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6441738480, "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['returned']:\n            continue\n        if r['region'] == 'Centre':\n            continue\n        if r['units'] < 4:\n            continue\n        total += r['units'] * r['unit_price'] * (1 - r['discount_pct'] / 100.0)\n    total *= 1.09975\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "b70b439dfd59fac1986bf92e0ca9882c51c32b2b48ad8dbc4ad1f85d39abf574", "output_hash": "5f25173e7b1a616c4eea703e50311df5ebdd9ac271873ce14a3700496d807d0c"}54{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "39260170f1472d5be66284da", "attempt_id": "516353cb7d74", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:50.359602Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6284932360}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7892, "cache_tensor_bytes": 36857088, "serialized_payload_bytes": 36891270, "side_information_bytes": 34182, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 19.15379089768976, "doc_prefill_ms": 202.10316707380116, "compress_ms": 10.761042009107769, "serialize_ms": 8.018375025130808, "load_ms": 2.049917005933821, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.807749941013753, "ttft_ms": 20.807749941013753, "decode_ms": 1510.7287920545787}, "total_wall_ms": 1535.2104170015082, "ttft_ms": 20.807749941013753, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 105}, "provider_cached_usage": null, "process_memory": 6929842176, "backend_memory": {"mlx_active_bytes": 6325020808, "mlx_peak_bytes": 6391149740, "mlx_cache_bytes": 4673725525, "process_rss_bytes": 6929842176}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6391149740, "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['returned']:\n            continue\n        if r['region'] == 'Centre':\n            continue\n        if r['units'] < 4:\n            continue\n        total += r['units'] * r['unit_price'] * (1 - r['discount_pct'] / 100.0)\n    total *= 1.09975\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "2a02a04b0b266a48baa8d83bddd6d77d388424ce5464dbbf336498026e33e76e", "output_hash": "5f25173e7b1a616c4eea703e50311df5ebdd9ac271873ce14a3700496d807d0c"}55{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "def96a8c91bcacab03e28d25", "attempt_id": "60dc6238f505", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:51.946008Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266101768}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1882, "keep": 470, "compress_ms": 8.438916062004864}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7892, "cache_tensor_bytes": 17326080, "serialized_payload_bytes": 17341989, "side_information_bytes": 15909, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 13.212124933488667, "doc_prefill_ms": 202.10316707380116, "compress_ms": 8.475666050799191, "serialize_ms": 4.556125029921532, "load_ms": 0.9394999360665679, "receiver_prefill_ms": 0.0, "query_prefill_ms": 18.26412498485297, "ttft_ms": 18.26412498485297, "decode_ms": 831.1934999655932}, "total_wall_ms": 852.0092910621315, "ttft_ms": 18.26412498485297, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 62}, "provider_cached_usage": null, "process_memory": 6947487744, "backend_memory": {"mlx_active_bytes": 6287400968, "mlx_peak_bytes": 6328803568, "mlx_cache_bytes": 4776291541, "process_rss_bytes": 6947487744}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6328803568, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n    total_revenue = 0\n    for row in rows:\n        if row['status'] == 'completed' and row['type'] == 'sale':\n            total_revenue += row['amount']\n    return total_revenue * (1 - row['tax_rate'])", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "107843b574e52ea6239922c9e359246508dfa59c5ee0c2c5e06c196ab15f2ddc", "output_hash": "e314a0b6bfae47115b5fa74c850b8faca507eac492d2e200e727f5820d8529c6"}56{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "945b15c978aa9a157157d18b", "attempt_id": "9054af90c199", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:52.836776Z", "task_family": "exec", "document_id": "exec_00201_2000", "document_hash": "065b6c18bbd16f16126924a54946ca4bf95c5863e57b14f33e4c9d88568f8549", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00201_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1882, "keep": 470, "window_tokens": 44, "compress_ms": 41.40049999114126}}, "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": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7892, "cache_tensor_bytes": 17326080, "serialized_payload_bytes": 17342029, "side_information_bytes": 15949, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 46.63254204206169, "doc_prefill_ms": 202.10316707380116, "compress_ms": 42.778416885994375, "serialize_ms": 3.646958968602121, "load_ms": 1.046125078573823, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.988165956921875, "ttft_ms": 17.988165956921875, "decode_ms": 1274.451583973132}, "total_wall_ms": 1341.6669580619782, "ttft_ms": 17.988165956921875, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 95}, "provider_cached_usage": null, "process_memory": 6950830080, "backend_memory": {"mlx_active_bytes": 6288809992, "mlx_peak_bytes": 6328148208, "mlx_cache_bytes": 4921748829, "process_rss_bytes": 6950830080}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6328148208, "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'] == 'Centre':\n            continue\n        if r['units'] < 4:\n            continue\n        if r['returned']:\n            continue\n        revenue = r['units'] * r['price'] * (1 + r['tax_rate'] / 100)\n        total += revenue\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00201_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/E1_core_repro.log", "payload_hash": "f4ac2f6c795ec165bce365482b0d2b69e5df048a1715190a5e2b2b6d6b901013", "output_hash": "13fb40b63bd476c948c0681cb512c968b7489f65d295ff3a52dffafb1eef7314"}57{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "0b3973eeba71c76cf21205d7", "attempt_id": "06793b1b6382", "run_group_id": "E1_core_repro:A1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:24:54.270001Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A1", "implementation_fidelity": "custom", "compression_config": {}, "question_access": "none", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": false, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 1926, "message_tokens": 1843, "message_utf8_bytes": 7665, "cache_tensor_bytes": null, "serialized_payload_bytes": 7665, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 0.008833943866193295, "decode_payload_ms": 0.0, "receiver_prefill_ms": 206.53924997895956, "ttft_ms": 13.80462502129376, "decode_ms": 1378.0347080901265}, "total_wall_ms": 1601.248958031647, "ttft_ms": 13.80462502129376, "input_usage": {"receiver_prompt_tokens": 1926}, "output_usage": {"output_tokens": 102}, "provider_cached_usage": null, "process_memory": 6589415424, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 7014645332, "mlx_cache_bytes": 851170252, "process_rss_bytes": 6589415424}, "peak_prefill_memory": null, "peak_decode_memory": 7014645332, "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'] == 'Est':\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    total *= 1.05\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00202_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/E1_core_repro.log", "payload_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "output_hash": "6263b0a9e86acb28188f4e95fe97726c1fa73d4518e84a7253d363b027a132cd"}58{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "e48ddf4e3667847e92e918b6", "attempt_id": "1bbff6b1ab20", "run_group_id": "E1_core_repro:A4|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:25:01.561332Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A4", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "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": 1843, "source_tokens_receiver": 1843, "query_tokens": 493, "message_tokens": 406, "message_utf8_bytes": 1776, "cache_tensor_bytes": null, "serialized_payload_bytes": 1776, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 5652.938166982494, "decode_payload_ms": 0.0, "receiver_prefill_ms": 56.1349579365924, "ttft_ms": 13.126457924954593, "decode_ms": 1246.0355840157717}, "total_wall_ms": 1316.471082973294, "ttft_ms": 13.126457924954593, "input_usage": {"receiver_prompt_tokens": 493, "sender_input_tokens": 1959, "sender_output_tokens": 406, "sender_prefill_ms": 209.25025001633912, "sender_ttft_ms": 14.038417022675276, "sender_decode_ms": 5427.731040981598}, "output_usage": {"output_tokens": 95}, "provider_cached_usage": null, "process_memory": 6589480960, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6494480392, "mlx_cache_bytes": 2154552188, "process_rss_bytes": 6589480960}, "peak_prefill_memory": null, "peak_decode_memory": 6494480392, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "```python\ndef net_revenue(rows):\n    total = 0.0\n    for r in rows:\n        if r['returned'] or r['region'] == 'Est' or r['units'] < 3:\n            continue\n        total += r['units'] * r['unit_price'] * (1 - r['discount_pct'] / 100.0)\n    return round(total * 1.05, 0)\n```", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00202_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/E1_core_repro.log", "payload_hash": "311a95c68f4ae9cc270cc34c02f8a63c4b03665e8b74ee3f462890b8dde1efeb", "output_hash": "90fa1ad446ffe0e1148fe5aa5978cff14a135616c6963cc7924255b05a1a705c"}59{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "58f5eabda03cf67c34857495", "attempt_id": "91e9a8ef7b18", "run_group_id": "E1_core_repro:A5|budget_tokens=512|blind", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:25:09.939474Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6171877384}, "weight_precision": "bf16", "method_id": "A5", "implementation_fidelity": "custom", "compression_config": {"access": "blind", "budget_tokens": 512, "realized": {"budget_tokens": 512, "n_facts": 7, "unresolved_quotes": 0, "parse_ok": 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": 1843, "source_tokens_receiver": 1843, "query_tokens": 483, "message_tokens": 384, "message_utf8_bytes": 1313, "cache_tensor_bytes": null, "serialized_payload_bytes": 1313, "side_information_bytes": 0, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 7024.673542007804, "decode_payload_ms": 0.0, "receiver_prefill_ms": 56.18129100184888, "ttft_ms": 13.417167006991804, "decode_ms": 1108.5099999327213}, "total_wall_ms": 1179.1344159282744, "ttft_ms": 13.417167006991804, "input_usage": {"receiver_prompt_tokens": 483, "sender_input_tokens": 1987, "sender_output_tokens": 512, "sender_prefill_ms": 210.3859999915585, "sender_ttft_ms": 13.527291943319142, "sender_decode_ms": 6798.74770808965}, "output_usage": {"output_tokens": 85}, "provider_cached_usage": null, "process_memory": 6589497344, "backend_memory": {"mlx_active_bytes": 6171877384, "mlx_peak_bytes": 6488680456, "mlx_cache_bytes": 3229981564, "process_rss_bytes": 6589497344}, "peak_prefill_memory": null, "peak_decode_memory": 6488680456, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n    rows = [row for row in rows if row['region'] != 'Est' and len(row['units']) >= 3]\n    revenue = sum(row['units'] * row['unit_price'] * (1 - row['discount_pct'] / 100) for row in rows)\n    revenue = revenue * 1.05\n    return round(revenue)", "evidence_artifact": {"n_facts": 7, "unresolved_quotes": 0, "parse_ok": true}, "reference_key_id": "keys/exec/exec_00202_2000.json", "task_scores": {"correct": 0.0, "pass_rate": 0.0, "error": "TypeError"}, "status": "success", "error_code": null, "retry_of": null, "log_artifact": "runs/logs/E1_core_repro.log", "payload_hash": "edd7f4a8c903d01a06f7424dde9759bab76ee3eecb0a4971eed2b80834da36fd", "output_hash": "d74f08cd5e7bdde4b2619af81be5e29233791ecbdf6e9ba0100808f63cbbdf78"}60{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "97d041282687a3324cdc5417", "attempt_id": "8602c5e29a89", "run_group_id": "E1_core_repro:B0", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:25:11.398797Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6316974088}, "weight_precision": "bf16", "method_id": "B0", "implementation_fidelity": "custom", "compression_config": {"realized": {}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7665, "cache_tensor_bytes": 69378048, "serialized_payload_bytes": 69394120, "side_information_bytes": 16072, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 242.398125003092, "doc_prefill_ms": 224.433625000529, "compress_ms": 1.7935410141944885, "serialize_ms": 12.897124979645014, "load_ms": 1.9019169267266989, "receiver_prefill_ms": 0.0, "query_prefill_ms": 30.687332968227565, "ttft_ms": 30.687332968227565, "decode_ms": 1428.62404207699}, "total_wall_ms": 1462.7205420983955, "ttft_ms": 30.687332968227565, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 102}, "provider_cached_usage": null, "process_memory": 6728515584, "backend_memory": {"mlx_active_bytes": 6392471560, "mlx_peak_bytes": 6441935088, "mlx_cache_bytes": 4239301853, "process_rss_bytes": 6728515584}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6441935088, "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'] == 'Est':\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    total *= 1.05\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00202_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/E1_core_repro.log", "payload_hash": "3036b65d7b530e49ba6e53370a2704f0f4894b07606a6893703eda3bfbd60971", "output_hash": "6263b0a9e86acb28188f4e95fe97726c1fa73d4518e84a7253d363b027a132cd"}61{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "258d0996a35e9824cf3ebf68", "attempt_id": "3041cc20686e", "run_group_id": "E1_core_repro:B1", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:25:12.912261Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6316974088}, "weight_precision": "bf16", "method_id": "B1", "implementation_fidelity": "custom", "compression_config": {"realized": {"serialize_roundtrip": true}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": "f683b7adf572348e", "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7665, "cache_tensor_bytes": 69378048, "serialized_payload_bytes": 69394148, "side_information_bytes": 16100, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 13.462791917845607, "doc_prefill_ms": 224.433625000529, "compress_ms": 1.6479159239679575, "serialize_ms": 11.632375069893897, "load_ms": 7.001832942478359, "receiver_prefill_ms": 0.0, "query_prefill_ms": 27.409249916672707, "ttft_ms": 27.409249916672707, "decode_ms": 1391.0934999585152}, "total_wall_ms": 1427.068084012717, "ttft_ms": 27.409249916672707, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 102}, "provider_cached_usage": null, "process_memory": 6798114816, "backend_memory": {"mlx_active_bytes": 6392471560, "mlx_peak_bytes": 6441738480, "mlx_cache_bytes": 4239301853, "process_rss_bytes": 6798114816}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6441738480, "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'] == 'Est':\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    total *= 1.05\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00202_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/E1_core_repro.log", "payload_hash": "b70b439dfd59fac1986bf92e0ca9882c51c32b2b48ad8dbc4ad1f85d39abf574", "output_hash": "6263b0a9e86acb28188f4e95fe97726c1fa73d4518e84a7253d363b027a132cd"}62{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "6771f9fbf25d79efab985537", "attempt_id": "c5a4f78a1845", "run_group_id": "E1_core_repro:B2|bits=8", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:25:14.397258Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6284932360}, "weight_precision": "bf16", "method_id": "B2", "implementation_fidelity": "custom", "compression_config": {"bits": 8, "realized": {"bits": 8}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7665, "cache_tensor_bytes": 36857088, "serialized_payload_bytes": 36891270, "side_information_bytes": 34182, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 19.92066705133766, "doc_prefill_ms": 224.433625000529, "compress_ms": 11.175042018294334, "serialize_ms": 8.36129195522517, "load_ms": 1.9638750236481428, "receiver_prefill_ms": 0.0, "query_prefill_ms": 20.71762504056096, "ttft_ms": 20.71762504056096, "decode_ms": 1469.080084003508}, "total_wall_ms": 1493.2396250078455, "ttft_ms": 20.71762504056096, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 102}, "provider_cached_usage": null, "process_memory": 6835552256, "backend_memory": {"mlx_active_bytes": 6325017352, "mlx_peak_bytes": 6391157324, "mlx_cache_bytes": 4726849853, "process_rss_bytes": 6835552256}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6391157324, "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'] == 'Est':\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    total *= 1.05\n    return round(total, 0)", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00202_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/E1_core_repro.log", "payload_hash": "2a02a04b0b266a48baa8d83bddd6d77d388424ce5464dbbf336498026e33e76e", "output_hash": "6263b0a9e86acb28188f4e95fe97726c1fa73d4518e84a7253d363b027a132cd"}63{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "a51162a92ef8e06c5cea95b2", "attempt_id": "4ac833ee45b3", "run_group_id": "E1_core_repro:B3|keep_frac=0.25", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:25:15.943685Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6266249224}, "weight_precision": "bf16", "method_id": "B3", "implementation_fidelity": "inspired", "compression_config": {"keep_frac": 0.25, "realized": {"policy": "sink_recent", "keep_frac": 0.25, "T": 1882, "keep": 470, "compress_ms": 8.492292021401227}}, "question_access": "blind", "topology": "single_handoff", "handoff_index": 1, "reuse_count": 1, "cache_hit": true, "cache_compatibility_hash": null, "random_seed": 0, "source_tokens_sender": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7665, "cache_tensor_bytes": 17326080, "serialized_payload_bytes": 17341989, "side_information_bytes": 15909, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 14.354042010381818, "doc_prefill_ms": 224.433625000529, "compress_ms": 8.528208010829985, "serialize_ms": 5.641167052090168, "load_ms": 1.5238749329000711, "receiver_prefill_ms": 0.0, "query_prefill_ms": 18.73050001449883, "ttft_ms": 18.73050001449883, "decode_ms": 828.7190409610048}, "total_wall_ms": 850.4767080303282, "ttft_ms": 18.73050001449883, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 63}, "provider_cached_usage": null, "process_memory": 6849642496, "backend_memory": {"mlx_active_bytes": 6288662536, "mlx_peak_bytes": 6328951024, "mlx_cache_bytes": 4849320853, "process_rss_bytes": 6849642496}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6328951024, "mapper_memory": null, "cost_with_source": null, "prediction_artifact": "def net_revenue(rows):\n    total_revenue = 0\n    for row in rows:\n        if row['status'] == 'completed' and row['amount'] > 0:\n            total_revenue += row['amount']\n    return total_revenue * 1.1 - 100", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00202_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/E1_core_repro.log", "payload_hash": "107843b574e52ea6239922c9e359246508dfa59c5ee0c2c5e06c196ab15f2ddc", "output_hash": "ac22038302f4adad3447b3b21febc62f0a85d78fbb35610b2c5bac1efc4a29d2"}64{"schema_version": "wp30-attempt-1", "study_id": "WP30", "experiment_id": "E1_core_repro", "trial_id": "746eaea14ea85922e6eeb36a", "attempt_id": "cdef9a8e95cb", "run_group_id": "E1_core_repro:B5|keep_frac=0.25|aware", "protocol_hash": "404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6", "config_hash": "10d4fe922e5c14d964ec64c65177898169e6fef2d18de1a0144163cca2c8fdcc", "git_commit": "d40b392c23efff803ea24b0c3a963a6e8aab4a32", "timestamp_utc": "2026-10-30T06:25:16.833181Z", "task_family": "exec", "document_id": "exec_00202_2000", "document_hash": "ec13194237af7efe7765a21f1d8303ab8ac449271eb48e3395906648c770bab8", "question_id": "q00_exec", "split": "test", "cluster_id": "exec_00202_2000", "sender_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "receiver_model_revision": "mlx-community/Qwen2.5-3B-Instruct-bf16@bec1ad658e380044bbfe06d4a2268801d3379a11", "tokenizer_hashes": {"sender": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b", "receiver": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b"}, "backend_versions": {"mlx": "0.32.4", "mlx_lm": "0.32.2"}, "hardware_id": "laptop-m5max-48g", "device_state": {"mlx_active_before": 6247374856}, "weight_precision": "bf16", "method_id": "B5", "implementation_fidelity": "faithful_reimplementation", "compression_config": {"access": "aware", "keep_frac": 0.25, "realized": {"policy": "attention", "keep_frac": 0.25, "T": 1882, "keep": 470, "window_tokens": 44, "compress_ms": 41.95012501440942}}, "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": 1843, "source_tokens_receiver": 1843, "query_tokens": 44, "message_tokens": 1882, "message_utf8_bytes": 7665, "cache_tensor_bytes": 17326080, "serialized_payload_bytes": 17342029, "side_information_bytes": 15949, "fetched_bytes": 0, "transport_bytes_if_measured": null, "timing_spans": {"encode_ms": 49.026416963897645, "doc_prefill_ms": 224.433625000529, "compress_ms": 43.287709006108344, "serialize_ms": 5.51979208830744, "load_ms": 0.9834590600803494, "receiver_prefill_ms": 0.0, "query_prefill_ms": 17.908332985825837, "ttft_ms": 17.908332985825837, "decode_ms": 1053.9388329489157}, "total_wall_ms": 1123.4697500476614, "ttft_ms": 17.908332985825837, "input_usage": {"receiver_prefix_tokens": 1882, "dense_cache_bytes": 69378048, "query_tokens": 44}, "output_usage": {"output_tokens": 80}, "provider_cached_usage": null, "process_memory": 6854983680, "backend_memory": {"mlx_active_bytes": 6287777800, "mlx_peak_bytes": 6328295664, "mlx_cache_bytes": 4954244125, "process_rss_bytes": 6854983680}, "peak_prefill_memory": 6997163604, "peak_decode_memory": 6328295664, "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'] != 'Est' and r['units'] >= 3 and not r['returned']:\n            revenue = r['line_revenue'] * 1.05  # Apply 5% tax\n            total += round(revenue, 0)\n    return total", "evidence_artifact": null, "reference_key_id": "keys/exec/exec_00202_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/E1_core_repro.log", "payload_hash": "f4ac2f6c795ec165bce365482b0d2b69e5df048a1715190a5e2b2b6d6b901013", "output_hash": "a0dfbedef433f3fe7a2e09f0c29f2e2ac714057fd5d22c4b3d0e599eb11ef447"}65