{
  "available": true,
  "version": "competition500-direct-review-evaluation-v3-20261019",
  "scope": "closed_500_cases_directly_reviewed_reference",
  "selected_mode": "deepseek_adaptive",
  "default_config": {
    "rerank_provider": "deepseek",
    "retrieval_timing": "adaptive",
    "query_anchors": 0,
    "retrieval_top_k": 5,
    "local_calibration": true,
    "longtext_screening": true
  },
  "dataset": {
    "version": "competition500-codex-reviewed-v2-20261019",
    "cases": 500,
    "source_documents": 500,
    "development_queries": 400,
    "development_validation_queries": 100,
    "test_queries": 200,
    "label_count": 27,
    "scene_count": 6,
    "annotation": "Codex逐条自复核",
    "human_expert_gold": false,
    "qrels_protocol": "direct-source-review-mechanisms-v3",
    "explicit_test_query_case_pairs": 100000,
    "test_used_for_training_or_selection": false,
    "external_model_review_calls": 0
  },
  "risk": {
    "deepseek_adaptive": {
      "queries": 200,
      "state_accuracy": 0.875,
      "risk_binary_accuracy": 0.91,
      "risk_precision": 0.8770491803278688,
      "risk_recall": 0.9727272727272728,
      "risk_f1": 0.9224137931034483,
      "scene_accuracy": 0.95,
      "label_micro_f1": 0.6500655307994757,
      "label_macro_f1": 0.6595567984003188,
      "label_supported_macro_f1": 0.6595567984003188,
      "per_label": {
        "欺骗投保人": {
          "tp": 6,
          "fp": 19,
          "fn": 1,
          "f1": 0.375,
          "support": 7
        },
        "销售误导": {
          "tp": 11,
          "fp": 26,
          "fn": 4,
          "f1": 0.4230769230769231,
          "support": 15
        },
        "夸大收益": {
          "tp": 15,
          "fp": 5,
          "fn": 1,
          "f1": 0.8333333333333334,
          "support": 16
        },
        "虚假宣传": {
          "tp": 12,
          "fp": 7,
          "fn": 4,
          "f1": 0.6857142857142857,
          "support": 16
        },
        "合同外利益": {
          "tp": 4,
          "fp": 7,
          "fn": 1,
          "f1": 0.5,
          "support": 5
        },
        "赠送利益": {
          "tp": 19,
          "fp": 3,
          "fn": 0,
          "f1": 0.926829268292683,
          "support": 19
        },
        "回扣返佣": {
          "tp": 4,
          "fp": 4,
          "fn": 2,
          "f1": 0.5714285714285714,
          "support": 6
        },
        "产品说明会违规": {
          "tp": 13,
          "fp": 3,
          "fn": 6,
          "f1": 0.7428571428571429,
          "support": 19
        },
        "培训材料违规": {
          "tp": 16,
          "fp": 10,
          "fn": 1,
          "f1": 0.7441860465116279,
          "support": 17
        },
        "回访违规": {
          "tp": 4,
          "fp": 0,
          "fn": 2,
          "f1": 0.8,
          "support": 6
        },
        "电销违规": {
          "tp": 7,
          "fp": 2,
          "fn": 1,
          "f1": 0.8235294117647058,
          "support": 8
        },
        "不当比较": {
          "tp": 12,
          "fp": 2,
          "fn": 1,
          "f1": 0.8888888888888888,
          "support": 13
        },
        "贬低竞品": {
          "tp": 4,
          "fp": 2,
          "fn": 0,
          "f1": 0.8,
          "support": 4
        },
        "绝对化表述": {
          "tp": 9,
          "fp": 4,
          "fn": 2,
          "f1": 0.75,
          "support": 11
        },
        "承诺收益": {
          "tp": 12,
          "fp": 6,
          "fn": 2,
          "f1": 0.75,
          "support": 14
        },
        "保证收益": {
          "tp": 12,
          "fp": 1,
          "fn": 2,
          "f1": 0.8888888888888888,
          "support": 14
        },
        "弱化风险提示": {
          "tp": 6,
          "fp": 2,
          "fn": 8,
          "f1": 0.5454545454545454,
          "support": 14
        },
        "隐瞒重要信息": {
          "tp": 6,
          "fp": 25,
          "fn": 1,
          "f1": 0.3157894736842105,
          "support": 7
        },
        "虚构收益率": {
          "tp": 1,
          "fp": 4,
          "fn": 0,
          "f1": 0.3333333333333333,
          "support": 1
        },
        "诱导投保": {
          "tp": 3,
          "fp": 17,
          "fn": 0,
          "f1": 0.2608695652173913,
          "support": 3
        },
        "避债避税": {
          "tp": 2,
          "fp": 0,
          "fn": 0,
          "f1": 1.0,
          "support": 2
        },
        "虚列费用套取资金": {
          "tp": 26,
          "fp": 17,
          "fn": 1,
          "f1": 0.7428571428571429,
          "support": 27
        },
        "代理人管理不到位": {
          "tp": 7,
          "fp": 4,
          "fn": 4,
          "f1": 0.6363636363636364,
          "support": 11
        },
        "委托无资质机构销售": {
          "tp": 2,
          "fp": 4,
          "fn": 0,
          "f1": 0.5,
          "support": 2
        },
        "未按规定使用费率条款": {
          "tp": 14,
          "fp": 2,
          "fn": 0,
          "f1": 0.9333333333333333,
          "support": 14
        },
        "客户信息不真实": {
          "tp": 9,
          "fp": 9,
          "fn": 1,
          "f1": 0.6428571428571429,
          "support": 10
        },
        "其他": {
          "tp": 12,
          "fp": 6,
          "fn": 31,
          "f1": 0.39344262295081966,
          "support": 43
        }
      },
      "median_seconds": 4.428342500003055,
      "p95_seconds": 11.781458300014492,
      "warnings": 1,
      "errors": 0,
      "missing_model_targets": 0,
      "long_queries": 1,
      "long_median_seconds": 6.056131299992558,
      "label_micro_precision": 0.5649202733485194,
      "label_micro_recall": 0.7654320987654321,
      "label_counts": {
        "tp": 248,
        "fp": 191,
        "fn": 76
      },
      "exact_label_set_accuracy": 0.51
    }
  },
  "risk_names": {
    "deepseek_adaptive": "DeepSeek · 自适应融合"
  },
  "retrieval": {
    "deepseek": {
      "judged_queries": 200,
      "positive_queries": 187,
      "no_positive_queries": 13,
      "hit1": 0.8021390374331551,
      "hit5": 0.9251336898395722,
      "mrr": 0.853223989320246,
      "recall_at_3": 0.12110474598402513,
      "recall_at_5": 0.16555648781948323,
      "recall_at_10": 0.236239739788865,
      "recall_at_20": 0.30487051842519364,
      "ndcg_at_10": 0.5818025039957844,
      "median_seconds": 1.2838264500314835
    }
  },
  "retrieval_measurement_note": "按逐条复核的具体行为目录与来源证据重新评测。",
  "review": {
    "query_rows_changed": 357,
    "case_rows_corrected": 72,
    "context_sources_rechecked": 33,
    "test_predictions_used_for_gold": false
  },
  "length_bands": {
    "under500": {
      "queries": 178,
      "risk_binary_accuracy": 0.9213483146067416,
      "risk_recall": 0.979381443298969,
      "label_micro_f1": 0.6475903614457831,
      "median_seconds": 4.114432950009359,
      "p95_seconds": 10.171660699998029,
      "errors": 0,
      "missing_model_targets": 0,
      "gold_risk_queries": 97,
      "label_support": 284,
      "false_positive_queries": 12
    },
    "500to899": {
      "queries": 21,
      "risk_binary_accuracy": 0.8571428571428571,
      "risk_recall": 0.9230769230769231,
      "label_micro_f1": 0.6875,
      "median_seconds": 6.450408199976664,
      "p95_seconds": 19.663018899969757,
      "errors": 0,
      "missing_model_targets": 0,
      "gold_risk_queries": 13,
      "label_support": 40,
      "false_positive_queries": 2
    },
    "900plus": {
      "queries": 1,
      "risk_binary_accuracy": 0.0,
      "risk_recall": 0.0,
      "label_micro_f1": 0.0,
      "median_seconds": 6.056131299992558,
      "p95_seconds": 6.056131299992558,
      "errors": 0,
      "missing_model_targets": 0,
      "gold_risk_queries": 0,
      "label_support": 0,
      "false_positive_queries": 1
    }
  },
  "provenance": {
    "run": "self_review_final_v3",
    "label_reference_sha256": "40c2fb580b6ae3b24ef781fdc7eaabbeaab4290e88de83ab8a77f39168be8b70",
    "test_input_sha256": "af6e34059a655ed16169b09357cda4e21e353cacac20bcf29e073f7eb3f18cfd",
    "test_references_modified_for_this_upgrade": true,
    "labels_modified_using_predictions": false,
    "external_model_review_calls": 0,
    "comparison_note": "Reference changed; old silver-reference numbers are not comparable.",
    "risk_api_tasks": {
      "risk_review": 218,
      "scene_classification": 28
    },
    "risk_api_errors": 1,
    "all_ranked_evidence_verified": true
  },
  "previous_predictions_rescored": {
    "source_run": "label_precision_v2",
    "reference": "competition500-codex-reviewed-v2-20261019",
    "new_api_run_of_previous_model": false,
    "metrics": {
      "queries": 200,
      "state_accuracy": 0.83,
      "risk_binary_accuracy": 0.88,
      "risk_precision": 0.8359375,
      "risk_recall": 0.9727272727272728,
      "risk_f1": 0.8991596638655462,
      "scene_accuracy": 0.905,
      "label_micro_f1": 0.622991347342398,
      "label_macro_f1": 0.6542602947349703,
      "label_supported_macro_f1": 0.6542602947349703,
      "per_label": {
        "欺骗投保人": {
          "tp": 6,
          "fp": 24,
          "fn": 1,
          "f1": 0.32432432432432434,
          "support": 7
        },
        "销售误导": {
          "tp": 13,
          "fp": 35,
          "fn": 2,
          "f1": 0.4126984126984127,
          "support": 15
        },
        "夸大收益": {
          "tp": 15,
          "fp": 5,
          "fn": 1,
          "f1": 0.8333333333333334,
          "support": 16
        },
        "虚假宣传": {
          "tp": 13,
          "fp": 10,
          "fn": 3,
          "f1": 0.6666666666666666,
          "support": 16
        },
        "合同外利益": {
          "tp": 5,
          "fp": 11,
          "fn": 0,
          "f1": 0.47619047619047616,
          "support": 5
        },
        "赠送利益": {
          "tp": 19,
          "fp": 4,
          "fn": 0,
          "f1": 0.9047619047619048,
          "support": 19
        },
        "回扣返佣": {
          "tp": 6,
          "fp": 5,
          "fn": 0,
          "f1": 0.7058823529411765,
          "support": 6
        },
        "产品说明会违规": {
          "tp": 11,
          "fp": 1,
          "fn": 8,
          "f1": 0.7096774193548387,
          "support": 19
        },
        "培训材料违规": {
          "tp": 16,
          "fp": 14,
          "fn": 1,
          "f1": 0.6808510638297872,
          "support": 17
        },
        "回访违规": {
          "tp": 5,
          "fp": 0,
          "fn": 1,
          "f1": 0.9090909090909091,
          "support": 6
        },
        "电销违规": {
          "tp": 7,
          "fp": 1,
          "fn": 1,
          "f1": 0.875,
          "support": 8
        },
        "不当比较": {
          "tp": 10,
          "fp": 4,
          "fn": 3,
          "f1": 0.7407407407407407,
          "support": 13
        },
        "贬低竞品": {
          "tp": 4,
          "fp": 2,
          "fn": 0,
          "f1": 0.8,
          "support": 4
        },
        "绝对化表述": {
          "tp": 10,
          "fp": 5,
          "fn": 1,
          "f1": 0.7692307692307693,
          "support": 11
        },
        "承诺收益": {
          "tp": 12,
          "fp": 6,
          "fn": 2,
          "f1": 0.75,
          "support": 14
        },
        "保证收益": {
          "tp": 11,
          "fp": 2,
          "fn": 3,
          "f1": 0.8148148148148148,
          "support": 14
        },
        "弱化风险提示": {
          "tp": 9,
          "fp": 6,
          "fn": 5,
          "f1": 0.6206896551724138,
          "support": 14
        },
        "隐瞒重要信息": {
          "tp": 6,
          "fp": 18,
          "fn": 1,
          "f1": 0.3870967741935484,
          "support": 7
        },
        "虚构收益率": {
          "tp": 1,
          "fp": 4,
          "fn": 0,
          "f1": 0.3333333333333333,
          "support": 1
        },
        "诱导投保": {
          "tp": 3,
          "fp": 22,
          "fn": 0,
          "f1": 0.21428571428571427,
          "support": 3
        },
        "避债避税": {
          "tp": 2,
          "fp": 0,
          "fn": 0,
          "f1": 1.0,
          "support": 2
        },
        "虚列费用套取资金": {
          "tp": 25,
          "fp": 17,
          "fn": 2,
          "f1": 0.7246376811594203,
          "support": 27
        },
        "代理人管理不到位": {
          "tp": 6,
          "fp": 9,
          "fn": 5,
          "f1": 0.46153846153846156,
          "support": 11
        },
        "委托无资质机构销售": {
          "tp": 2,
          "fp": 2,
          "fn": 0,
          "f1": 0.6666666666666666,
          "support": 2
        },
        "未按规定使用费率条款": {
          "tp": 14,
          "fp": 0,
          "fn": 0,
          "f1": 1.0,
          "support": 14
        },
        "客户信息不真实": {
          "tp": 9,
          "fp": 16,
          "fn": 1,
          "f1": 0.5142857142857142,
          "support": 10
        },
        "其他": {
          "tp": 12,
          "fp": 10,
          "fn": 31,
          "f1": 0.36923076923076925,
          "support": 43
        }
      },
      "median_seconds": 4.6253839999990305,
      "p95_seconds": 10.782025699998485,
      "warnings": 1,
      "errors": 0,
      "missing_model_targets": 0,
      "long_queries": 1,
      "long_median_seconds": 4.731980200012913
    }
  }
}