{ "model": "D1A-E4B v0.6", "temperature": 1.954, "method": "d1a.eval.benchmark through scripts/quality_gate.py and scripts/decide.py on the MLX 8-bit builds of v0.5 (T 1.782) and v0.6 (T 1.954), held-out partitions; paired bootstrap over questions (or PRs), 95% intervals. The fresh-PR set is the sealed EVAL half of hermes-agent PRs created after all training, scored with the serving context; its text is private.", "verdict_vs_v0.5": { "decide": "VETO: V1", "pooled": { "delta": -0.2, "ci95": [ -0.98, 0.52 ], "lines": 21 }, "release": "shipped as an explicit exception (see provenance.json)" }, "sets": { "devtools-v1": { "n": 1074, "v0.5": { "acc": 69.8, "ece": 0.048 }, "v0.6": { "acc": 70.1, "ece": 0.037 }, "delta": 0.3, "ci95": [ -1.4, 1.9 ], "fixes_breaks": [ 53, 50 ] }, "documents-v1": { "n": 920, "v0.5": { "acc": 87.5, "ece": 0.042 }, "v0.6": { "acc": 89.5, "ece": 0.028 }, "delta": 2.0, "ci95": [ 0.1, 3.7 ], "fixes_breaks": [ 43, 25 ] }, "external_semif-v1": { "n": 144, "v0.5": { "acc": 84.7, "ece": 0.11 }, "v0.6": { "acc": 83.3, "ece": 0.095 }, "delta": -1.4, "ci95": [ -4.9, 2.1 ], "fixes_breaks": [ 3, 5 ] }, "external_typesafe-v1": { "n": 102, "v0.5": { "acc": 74.5, "ece": 0.104 }, "v0.6": { "acc": 74.5, "ece": 0.129 }, "delta": 0.0, "ci95": [ -6.3, 7.1 ], "fixes_breaks": [ 7, 7 ] }, "external_wanli-v1": { "n": 256, "v0.5": { "acc": 72.7, "ece": 0.097 }, "v0.6": { "acc": 70.3, "ece": 0.101 }, "delta": -2.3, "ci95": [ -5.5, 0.8 ], "fixes_breaks": [ 6, 12 ] }, "external_wanli-v2": { "n": 1002, "v0.5": { "acc": 67.1, "ece": 0.123 }, "v0.6": { "acc": 67.1, "ece": 0.072 }, "delta": 0.0, "ci95": [ -1.9, 1.8 ], "fixes_breaks": [ 47, 47 ] }, "hard-v1": { "n": 1083, "v0.5": { "acc": 71.5, "ece": 0.045 }, "v0.6": { "acc": 68.8, "ece": 0.043 }, "delta": -2.7, "ci95": [ -4.9, -0.5 ], "fixes_breaks": [ 61, 90 ] }, "ja-jglue_development": { "n": 1500, "v0.5": { "acc": 82.1, "ece": 0.026 }, "v0.6": { "acc": 81.7, "ece": 0.042 }, "delta": -0.3, "ci95": [ -1.6, 0.9 ], "fixes_breaks": [ 42, 47 ] }, "ja-jglue_test": { "n": 1500, "v0.5": { "acc": 80.2, "ece": 0.024 }, "v0.6": { "acc": 80.5, "ece": 0.039 }, "delta": 0.3, "ci95": [ -0.9, 1.5 ], "fixes_breaks": [ 51, 46 ] }, "night2_assertion": { "n": 1020, "v0.5": { "acc": 94.4, "ece": 0.051 }, "v0.6": { "acc": 94.9, "ece": 0.056 }, "delta": 0.5, "ci95": [ -0.4, 1.4 ], "fixes_breaks": [ 12, 7 ] }, "night2_dates": { "n": 900, "v0.5": { "acc": 92.7, "ece": 0.028 }, "v0.6": { "acc": 92.1, "ece": 0.053 }, "delta": -0.6, "ci95": [ -2.2, 1.1 ], "fixes_breaks": [ 36, 41 ] }, "night2_unknowable": { "n": 525, "v0.5": { "acc": 68.2, "ece": 0.062 }, "v0.6": { "acc": 67.8, "ece": 0.058 }, "delta": -0.4, "ci95": [ -2.7, 1.7 ], "fixes_breaks": [ 30, 32 ] }, "pr-labels_development": { "n": 2030, "v0.5": { "acc": 80.9, "ece": 0.02 }, "v0.6": { "acc": 79.7, "ece": 0.017 }, "delta": -1.3, "ci95": [ -2.6, 0.0 ], "fixes_breaks": [ 80, 106 ] }, "pr-labels_development-ja": { "n": 198, "v0.5": { "acc": 77.3, "ece": 0.064 }, "v0.6": { "acc": 76.8, "ece": 0.053 }, "delta": -0.5, "ci95": [ -4.1, 3.1 ], "fixes_breaks": [ 6, 7 ] }, "pr-labels_real17": { "n": 51, "v0.5": { "acc": 64.7, "ece": 0.136 }, "v0.6": { "acc": 68.6, "ece": 0.181 }, "delta": 3.9, "ci95": [ 0.0, 9.8 ], "fixes_breaks": [ 2, 0 ] }, "pr-labels_test": { "n": 2045, "v0.5": { "acc": 81.3, "ece": 0.038 }, "v0.6": { "acc": 80.2, "ece": 0.03 }, "delta": -1.1, "ci95": [ -2.5, 0.4 ], "fixes_breaks": [ 106, 128 ] }, "pr-labels_test-ja": { "n": 193, "v0.5": { "acc": 76.2, "ece": 0.071 }, "v0.6": { "acc": 72.5, "ece": 0.043 }, "delta": -3.6, "ci95": [ -8.1, 0.5 ], "fixes_breaks": [ 6, 13 ] }, "routing_factory-development": { "n": 640, "v0.5": { "acc": 90.5, "ece": 0.13 }, "v0.6": { "acc": 93.9, "ece": 0.091 }, "delta": 3.4, "ci95": [ 1.4, 5.6 ], "fixes_breaks": [ 30, 8 ] }, "routing_generic-development": { "n": 270, "v0.5": { "acc": 98.9, "ece": 0.029 }, "v0.6": { "acc": 98.5, "ece": 0.015 }, "delta": -0.4, "ci95": [ -1.5, 0.7 ], "fixes_breaks": [ 1, 2 ] }, "routing_handlabelled-45": { "n": 45, "v0.5": { "acc": 100.0, "ece": 0.032 }, "v0.6": { "acc": 97.8, "ece": 0.014 }, "delta": -2.2, "ci95": [ -6.7, 0.0 ], "fixes_breaks": [ 0, 1 ] }, "v4_transfer-v4": { "n": 656, "v0.5": { "acc": 71.6, "ece": 0.106 }, "v0.6": { "acc": 73.0, "ece": 0.055 }, "delta": 1.4, "ci95": [ -1.2, 4.0 ], "fixes_breaks": [ 32, 23 ] }, "v7_decision-v7": { "n": 1264, "v0.5": { "acc": 85.3, "ece": 0.03 }, "v0.6": { "acc": 83.9, "ece": 0.015 }, "delta": -1.4, "ci95": [ -2.9, 0.0 ], "fixes_breaks": [ 31, 49 ] }, "report-only:transfer-v2": { "n": 560, "v0.5": { "acc": 74.1, "ece": 0.074 }, "v0.6": { "acc": 73.8, "ece": 0.041 }, "delta": -0.4, "ci95": [ -2.3, 1.6 ], "fixes_breaks": [ 17, 19 ], "report_only": true }, "report-only:v9_transfer-v9": { "n": 1046, "v0.5": { "acc": 65.3, "ece": 0.083 }, "v0.6": { "acc": 65.9, "ece": 0.054 }, "delta": 0.6, "ci95": [ -1.3, 2.4 ], "fixes_breaks": [ 49, 43 ], "report_only": true } }, "pr_label_questions": { "pr-labels_test:type": { "n": 953, "v0.5": 89.4, "v0.6": 85.3, "delta": -4.1, "ci95": [ -6.0, -2.3 ] }, "pr-labels_test:blast": { "n": 139, "v0.5": 57.6, "v0.6": 61.9, "delta": 4.3, "ci95": [ -2.2, 10.8 ] }, "pr-labels_test:sev": { "n": 953, "v0.5": 76.7, "v0.6": 77.9, "delta": 1.2, "ci95": [ -1.3, 3.6 ] }, "pr-labels_development:type": { "n": 946, "v0.5": 90.8, "v0.6": 90.7, "delta": -0.1, "ci95": [ -1.8, 1.5 ] }, "pr-labels_development:blast": { "n": 138, "v0.5": 58.7, "v0.6": 52.9, "delta": -5.8, "ci95": [ -11.6, 0.0 ] }, "pr-labels_development:sev": { "n": 946, "v0.5": 74.3, "v0.6": 72.5, "delta": -1.8, "ci95": [ -3.9, 0.3 ] }, "pr-labels_test-ja:type": { "n": 92, "v0.5": 83.7, "v0.6": 79.3, "delta": -4.3, "ci95": [ -9.8, 0.0 ] }, "pr-labels_test-ja:blast": { "n": 9, "v0.5": 44.4, "v0.6": 44.4, "delta": 0.0, "ci95": [ 0.0, 0.0 ] }, "pr-labels_test-ja:sev": { "n": 92, "v0.5": 71.7, "v0.6": 68.5, "delta": -3.3, "ci95": [ -10.9, 4.3 ] }, "pr-labels_development-ja:type": { "n": 95, "v0.5": 87.4, "v0.6": 85.3, "delta": -2.1, "ci95": [ -6.3, 1.1 ] }, "pr-labels_development-ja:blast": { "n": 8, "v0.5": 62.5, "v0.6": 50.0, "delta": -12.5, "ci95": [ -37.5, 0.0 ] }, "pr-labels_development-ja:sev": { "n": 95, "v0.5": 68.4, "v0.6": 70.5, "delta": 2.1, "ci95": [ -4.2, 7.4 ] }, "pr-labels_real17:type": { "n": 17, "v0.5": 82.4, "v0.6": 82.4, "delta": 0.0, "ci95": [ 0.0, 0.0 ] }, "pr-labels_real17:blast": { "n": 17, "v0.5": 58.8, "v0.6": 70.6, "delta": 11.8, "ci95": [ 0.0, 29.4 ] }, "pr-labels_real17:sev": { "n": 17, "v0.5": 52.9, "v0.6": 52.9, "delta": 0.0, "ci95": [ 0.0, 0.0 ] }, "pr-labels_test:sev+offsets": { "n": 953, "v0.5": 77.0, "v0.6": 76.6, "delta": -0.4, "ci95": [ -2.4, 1.8 ] } }, "fresh_prs_eval": { "sev": { "n": 487, "majority_floor": 0.678, "acc": { "v0.4": 0.774, "v0.5": 0.747, "v0.6": 0.791 }, "v0.6_vs_v0.5": { "delta": 4.3, "ci95": [ 0.8, 7.8 ], "fixes": 52, "breaks": 31, "sign_p": 0.028 }, "v0.6_vs_v0.4": { "delta": 1.6, "ci95": [ -0.8, 4.1 ] }, "recall": { "v0.4": { "P0": 0.29, "P1": 0.35, "P2": 0.58, "P3": 0.89, "P4": 0.5 }, "v0.5": { "P0": 0.29, "P1": 0.22, "P2": 0.79, "P3": 0.78, "P4": 0.5 }, "v0.6": { "P0": 0.71, "P1": 0.35, "P2": 0.55, "P3": 0.92, "P4": 0.5 } } }, "type": { "n": 487, "majority_floor": 0.515, "acc": { "v0.4": 0.864, "v0.5": 0.867, "v0.6": 0.906 }, "v0.6_vs_v0.5": { "delta": 3.9, "ci95": [ 1.2, 6.6 ], "fixes": 31, "breaks": 12, "sign_p": 0.005 }, "v0.6_vs_v0.4": { "delta": 4.1, "ci95": [ 1.4, 7.0 ] } } } }