{
  "published_at": "2026-10-01T00:52:32+08:00",
  "production_version_at_snapshot": "4.10.4",
  "completed_study": {
    "version": "4.10.0",
    "commit": "bb11baefc4d0dac5ac662263d90a2e8283ba30bc",
    "source_summary_sha256": "15780e14e2303adfc1cd6da23eeb64b63a4c644bb8e2f4e1d5147684b181fe7d",
    "turns": 16,
    "artifact_qa": {
      "pass": 16,
      "fail": 0,
      "unknown": 0
    },
    "tasks": {
      "pdf": 12,
      "volcano": 2,
      "synthetic_domain": 2
    },
    "automatic_review": {
      "pass": 10,
      "partial": 6
    },
    "recovered_tool_errors": 2,
    "pdf_round_timings_s": [
      {
        "requested_mode": "full",
        "repeat": 1,
        "turn": 1,
        "qualified_artifact_s": 2.713306,
        "first_text_s": 6.485333,
        "completion_s": 33.320136
      },
      {
        "requested_mode": "full",
        "repeat": 1,
        "turn": 2,
        "qualified_artifact_s": 2.6562379999999997,
        "first_text_s": 10.546437,
        "completion_s": 30.276897
      },
      {
        "requested_mode": "adaptive",
        "repeat": 1,
        "turn": 1,
        "qualified_artifact_s": 2.751207,
        "first_text_s": 6.044241,
        "completion_s": 24.597300999999998
      },
      {
        "requested_mode": "adaptive",
        "repeat": 1,
        "turn": 2,
        "qualified_artifact_s": 2.60362,
        "first_text_s": 6.11161,
        "completion_s": 25.523464
      },
      {
        "requested_mode": "adaptive",
        "repeat": 2,
        "turn": 1,
        "qualified_artifact_s": 2.919479,
        "first_text_s": 5.38791,
        "completion_s": 15.595531999999999
      },
      {
        "requested_mode": "adaptive",
        "repeat": 2,
        "turn": 2,
        "qualified_artifact_s": 2.206895,
        "first_text_s": 5.13448,
        "completion_s": 15.131397000000002
      },
      {
        "requested_mode": "full",
        "repeat": 2,
        "turn": 1,
        "qualified_artifact_s": 2.6439760000000003,
        "first_text_s": 6.493303,
        "completion_s": 19.28404
      },
      {
        "requested_mode": "full",
        "repeat": 2,
        "turn": 2,
        "qualified_artifact_s": 2.528559,
        "first_text_s": 3.827112,
        "completion_s": 12.55959
      },
      {
        "requested_mode": "full",
        "repeat": 3,
        "turn": 1,
        "qualified_artifact_s": 2.443078,
        "first_text_s": 11.178946,
        "completion_s": 32.892814
      },
      {
        "requested_mode": "full",
        "repeat": 3,
        "turn": 2,
        "qualified_artifact_s": 2.481441,
        "first_text_s": 7.148032,
        "completion_s": 30.917768
      },
      {
        "requested_mode": "adaptive",
        "repeat": 3,
        "turn": 1,
        "qualified_artifact_s": 2.743331,
        "first_text_s": 5.7804269999999995,
        "completion_s": 27.586239000000003
      },
      {
        "requested_mode": "adaptive",
        "repeat": 3,
        "turn": 2,
        "qualified_artifact_s": 2.449536,
        "first_text_s": 5.476416,
        "completion_s": 21.985967000000002
      }
    ],
    "offline_tests": {
      "pass": 1046,
      "fail": 0,
      "skipped": 8,
      "modules": 49
    },
    "environment": "macOS arm64 / local Linux runc; excludes production gateway",
    "models": {
      "main": "GLM-5.3-flashx",
      "review": "deepseek-v4-flash"
    },
    "limitations": [
      "PDF: 3 repeats per requested mode, 2 turns per repeat.",
      "Both requested modes use effective full for these PDF tasks.",
      "Natural cache was not cleared; timings are descriptive, not causal speed claims.",
      "Accepted artifact time is an observed tool-completion upper bound, confirmed by later QA.",
      "Formal batch excludes preserved historical failed candidates and prechecks.",
      "These checks do not establish complete scientific release-gate approval."
    ]
  },
  "synthetic_file_benchmark": {
    "repeats": 5,
    "statistic": "median",
    "unit": "ms",
    "cases": [
      {
        "label": "稳定大文件快照",
        "scope": "12 × 8 MiB",
        "before_ms": 26.137,
        "after_ms": 5.655
      },
      {
        "label": "稳定文件重复摘要",
        "scope": "16 MiB",
        "before_ms": 6.84,
        "after_ms": 0.116
      },
      {
        "label": "小文件快照",
        "scope": "100 × 8 KiB",
        "before_ms": 15.39,
        "after_ms": 39.928
      },
      {
        "label": "新文件摘要",
        "scope": "16 MiB",
        "before_ms": 6.132,
        "after_ms": 11.92
      }
    ],
    "source": "docs/PERFORMANCE_UPGRADE_20260930.md"
  },
  "release_evaluation_snapshot": {
    "version": "4.10.3",
    "commit": "939116d9ed5d76e76bd081d8b29d7b574437a48a",
    "published_release_version": "4.10.4",
    "bank_version": "2026-09-30.2",
    "completed_runs": 87,
    "completed_grades": 87,
    "planned_runs": 87,
    "questions": 29,
    "repeats": 3,
    "grader_version": "2026-09-30.3",
    "grader_source_commit": "bb64cc8ee32e4e1985e2c632e1f36224dbccfb86",
    "original_grading_gates": {
      "档均分": {
        "value": 85.25,
        "min": 70.0,
        "pass": true
      },
      "幻觉率": {
        "value": 0.0,
        "max": 0.0,
        "questions": [],
        "pass": true
      },
      "硬失败率": {
        "value": 0.1034,
        "max": 0.1,
        "questions": [
          "seqstruct-03",
          "seqstruct-04",
          "seqstruct-05"
        ],
        "pass": false
      },
      "交付完整率": {
        "value": 1.0,
        "min": 0.95,
        "pass": true
      }
    },
    "state": "passed",
    "gates": {
      "档均分": {
        "value": 85.62,
        "min": 70.0,
        "pass": true
      },
      "幻觉率": {
        "value": 0.0,
        "max": 0.0,
        "questions": [],
        "pass": true
      },
      "硬失败率": {
        "value": 0.0,
        "max": 0.1,
        "questions": [],
        "pass": true
      },
      "交付完整率": {
        "value": 1.0,
        "min": 0.95,
        "pass": true
      }
    },
    "snapshot_at": "2026-10-01T00:52:32+08:00",
    "mode": "full",
    "category_scores": [
      {
        "id": "reuse",
        "label": "公开数据复用",
        "questions": 6,
        "runs": 18,
        "mean_of_question_medians": 77.63
      },
      {
        "id": "seqstruct",
        "label": "序列与结构",
        "questions": 6,
        "runs": 18,
        "mean_of_question_medians": 87.27
      },
      {
        "id": "t2d",
        "label": "靶点与药物证据",
        "questions": 6,
        "runs": 18,
        "mean_of_question_medians": 74.36333333333334
      },
      {
        "id": "v11",
        "label": "扩展工具与交付",
        "questions": 11,
        "runs": 33,
        "mean_of_question_medians": 95.22272727272727
      }
    ],
    "question_scores": [
      {
        "qid": "reuse-01-islet-t2d",
        "runs": 3,
        "median": 75.4,
        "min": 72.3,
        "max": 77.86
      },
      {
        "qid": "reuse-02-lung-hlca",
        "runs": 3,
        "median": 84.33,
        "min": 76.47,
        "max": 88.29
      },
      {
        "qid": "reuse-03-retina-muller",
        "runs": 3,
        "median": 81.94,
        "min": 79.17,
        "max": 87.5
      },
      {
        "qid": "reuse-04-atlas-selection",
        "runs": 3,
        "median": 82.42,
        "min": 68.53,
        "max": 86.11
      },
      {
        "qid": "reuse-05-mouse-fallback",
        "runs": 3,
        "median": 82.88,
        "min": 75.0,
        "max": 84.72
      },
      {
        "qid": "reuse-06-preprint-honesty",
        "runs": 3,
        "median": 58.81,
        "min": 49.09,
        "max": 65.76
      },
      {
        "qid": "seqstruct-01",
        "runs": 3,
        "median": 88.57,
        "min": 85.12,
        "max": 93.57
      },
      {
        "qid": "seqstruct-02",
        "runs": 3,
        "median": 79.92,
        "min": 74.29,
        "max": 91.11
      },
      {
        "qid": "seqstruct-03",
        "runs": 3,
        "median": 85.48,
        "min": 81.59,
        "max": 86.35
      },
      {
        "qid": "seqstruct-04",
        "runs": 3,
        "median": 98.61,
        "min": 91.67,
        "max": 98.61
      },
      {
        "qid": "seqstruct-05",
        "runs": 3,
        "median": 77.98,
        "min": 76.59,
        "max": 81.35
      },
      {
        "qid": "seqstruct-06",
        "runs": 3,
        "median": 93.06,
        "min": 81.03,
        "max": 94.44
      },
      {
        "qid": "t2d-01",
        "runs": 3,
        "median": 78.7,
        "min": 70.85,
        "max": 81.02
      },
      {
        "qid": "t2d-02",
        "runs": 3,
        "median": 66.87,
        "min": 65.2,
        "max": 70.76
      },
      {
        "qid": "t2d-03",
        "runs": 3,
        "median": 64.29,
        "min": 60.12,
        "max": 64.29
      },
      {
        "qid": "t2d-04",
        "runs": 3,
        "median": 77.98,
        "min": 76.87,
        "max": 78.45
      },
      {
        "qid": "t2d-05",
        "runs": 3,
        "median": 93.06,
        "min": 92.14,
        "max": 93.06
      },
      {
        "qid": "t2d-06",
        "runs": 3,
        "median": 65.28,
        "min": 49.08,
        "max": 67.59
      },
      {
        "qid": "v11-01",
        "runs": 3,
        "median": 93.34,
        "min": 91.68,
        "max": 96.68
      },
      {
        "qid": "v11-02",
        "runs": 3,
        "median": 85.0,
        "min": 85.0,
        "max": 100.0
      },
      {
        "qid": "v11-03",
        "runs": 3,
        "median": 98.12,
        "min": 89.81,
        "max": 100.0
      },
      {
        "qid": "v11-04",
        "runs": 3,
        "median": 91.06,
        "min": 87.93,
        "max": 93.12
      },
      {
        "qid": "v11-05",
        "runs": 3,
        "median": 100.0,
        "min": 96.25,
        "max": 100.0
      },
      {
        "qid": "v11-06",
        "runs": 3,
        "median": 98.12,
        "min": 88.12,
        "max": 100.0
      },
      {
        "qid": "v11-07",
        "runs": 3,
        "median": 100.0,
        "min": 94.44,
        "max": 100.0
      },
      {
        "qid": "v11-08",
        "runs": 3,
        "median": 95.0,
        "min": 92.5,
        "max": 100.0
      },
      {
        "qid": "v11-09",
        "runs": 3,
        "median": 100.0,
        "min": 86.11,
        "max": 100.0
      },
      {
        "qid": "v11-10",
        "runs": 3,
        "median": 100.0,
        "min": 100.0,
        "max": 100.0
      },
      {
        "qid": "v11-11",
        "runs": 3,
        "median": 86.81,
        "min": 86.81,
        "max": 97.92
      }
    ],
    "runtime_review": {
      "pass": 82,
      "partial": 5
    },
    "step_limit_runs": 6,
    "source_grade_sha256": "b2d403a1f8003b49a02e23ab46581643726ba1503bca876a3c6a1ff9d0876704",
    "fresh_scientific_evaluation_for_4_10_4": false,
    "timing": {
      "p50_s": 502.861,
      "p90_s": 1094.218,
      "scope": "Local batch with six model threads and one executor lane; includes scheduling waits; not production response latency."
    }
  },
  "python_compatibility_release": {
    "version": "4.10.4",
    "commit": "4658c4bb8cc3101a311e8f4c98f70b3de5567349",
    "python_regression": {
      "version": "4.10.4",
      "source_commit": "4658c4bb8cc3101a311e8f4c98f70b3de5567349",
      "python": "3.10.21",
      "passed": 1060,
      "failed": 0,
      "skipped": 8,
      "modules": 51,
      "wall_s": 57.2,
      "first_attempt": {
        "passed": 1059,
        "failed": 1,
        "skipped": 8,
        "cause": "New worktree frontend dist had not been built; static-path test received503. Built the release frontend and repeated the entire Python suite."
      },
      "scientific_evaluation": "not rerun per explicit user instruction; reference unchanged v4.10.3 evidence"
    },
    "production_python310_checks": "passed",
    "fresh_scientific_evaluation": false,
    "evidence_basis": "v4.10.3"
  },
  "previous_failed_candidate": {
    "version": "4.10.2",
    "commit": "d35b77133dc5698732bfe2b2d5ad87cc1d0f21ca",
    "bank_version": "2026-09-30.1",
    "gates": {
      "档均分": {
        "value": 83.0,
        "min": 70.0,
        "pass": true
      },
      "幻觉率": {
        "value": 0.0,
        "max": 0.0,
        "questions": [],
        "pass": true
      },
      "硬失败率": {
        "value": 0.1034,
        "max": 0.1,
        "questions": [
          "t2d-01",
          "t2d-04",
          "v11-02"
        ],
        "pass": false
      },
      "交付完整率": {
        "value": 1.0,
        "min": 0.95,
        "pass": true
      }
    },
    "state": "failed"
  },
  "previous_failed_candidates": [
    {
      "version": "4.10.1",
      "commit": "061a8a7df4347ef3ef3680dde52989e3264832d5",
      "bank_version": "2026-08-28.1",
      "gates": {
        "档均分": {
          "value": 79.32,
          "min": 70.0,
          "pass": true
        },
        "幻觉率": {
          "value": 0.0,
          "max": 0.0,
          "questions": [],
          "pass": true
        },
        "硬失败率": {
          "value": 0.1379,
          "max": 0.1,
          "questions": [
            "t2d-04",
            "v11-08",
            "v11-10",
            "v11-11"
          ],
          "pass": false
        },
        "交付完整率": {
          "value": 1.0,
          "min": 0.95,
          "pass": true
        }
      },
      "state": "failed"
    },
    {
      "version": "4.10.2",
      "commit": "d35b77133dc5698732bfe2b2d5ad87cc1d0f21ca",
      "bank_version": "2026-09-30.1",
      "gates": {
        "档均分": {
          "value": 83.0,
          "min": 70.0,
          "pass": true
        },
        "幻觉率": {
          "value": 0.0,
          "max": 0.0,
          "questions": [],
          "pass": true
        },
        "硬失败率": {
          "value": 0.1034,
          "max": 0.1,
          "questions": [
            "t2d-01",
            "t2d-04",
            "v11-02"
          ],
          "pass": false
        },
        "交付完整率": {
          "value": 1.0,
          "min": 0.95,
          "pass": true
        }
      },
      "state": "failed"
    }
  ],
  "production_source_commit": "4658c4bb8cc3101a311e8f4c98f70b3de5567349",
  "deployment_date_beijing": "2026-10-01T00:23:48+08:00"
}