{
  "split": "dev",
  "input_root": "data_dpo_v2\\dev_pref_v1",
  "total_pairs": 300,
  "assembly_counts": {
    "chosen_rule_fallback": 154,
    "rejected_weak_model": 155,
    "rejected_constructed_fallback": 145,
    "chosen_strong_model": 146
  },
  "strong_answer_summary": {
    "total_episodes": 300,
    "detection_accuracy": 0.5466666666666666,
    "miss_rate": 0.41875,
    "false_alarm_rate": 0.4928571428571429,
    "precision": 0.5740740740740741,
    "recall": 0.58125,
    "f1_security": 0.577639751552795,
    "threat_type_accuracy": 0.8387096774193549,
    "parse_failure_rate": 0.016666666666666666,
    "breakdown": {
      "TP": {
        "total": 160,
        "correct": 93,
        "accuracy": 0.58125
      },
      "FP": {
        "total": 120,
        "correct": 56,
        "accuracy": 0.4666666666666667
      },
      "TN": {
        "total": 20,
        "correct": 15,
        "accuracy": 0.75
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 70,
        "accuracy": 0.5714285714285714,
        "miss_count": 2,
        "false_alarm_count": 28
      },
      "SQ2": {
        "total": 38,
        "accuracy": 0.5526315789473685,
        "miss_count": 13,
        "false_alarm_count": 4
      },
      "SQ3": {
        "total": 71,
        "accuracy": 0.5492957746478874,
        "miss_count": 23,
        "false_alarm_count": 9
      },
      "SQ4": {
        "total": 59,
        "accuracy": 0.559322033898305,
        "miss_count": 16,
        "false_alarm_count": 10
      },
      "SQ5": {
        "total": 62,
        "accuracy": 0.5,
        "miss_count": 13,
        "false_alarm_count": 18
      }
    },
    "label": "strong_answers",
    "count": 300,
    "error_taxonomy": {
      "total_errors": 151,
      "error_distribution": {
        "MISS": 65,
        "FALSE_ALARM": 69,
        "PARSE_FAIL": 2,
        "WRONG_TYPE": 15
      },
      "error_by_sq": {
        "SQ1": {
          "PARSE_FAIL": 2,
          "FALSE_ALARM": 28
        },
        "SQ2": {
          "MISS": 13,
          "FALSE_ALARM": 4
        },
        "SQ3": {
          "MISS": 23,
          "FALSE_ALARM": 9,
          "WRONG_TYPE": 9
        },
        "SQ4": {
          "FALSE_ALARM": 10,
          "WRONG_TYPE": 6,
          "MISS": 16
        },
        "SQ5": {
          "MISS": 13,
          "FALSE_ALARM": 18
        }
      },
      "error_by_category": {
        "intrusion": 44,
        "elderly_specific": 15,
        "child_specific": 11,
        "behavioral_anomaly": 10
      }
    }
  },
  "weak_answer_summary": {
    "total_episodes": 300,
    "detection_accuracy": 0.5966666666666667,
    "miss_rate": 0.29375,
    "false_alarm_rate": 0.5214285714285715,
    "precision": 0.6075268817204301,
    "recall": 0.70625,
    "f1_security": 0.653179190751445,
    "threat_type_accuracy": 0.7256637168141593,
    "parse_failure_rate": 0.02666666666666667,
    "breakdown": {
      "TP": {
        "total": 160,
        "correct": 113,
        "accuracy": 0.70625
      },
      "FP": {
        "total": 120,
        "correct": 48,
        "accuracy": 0.4
      },
      "TN": {
        "total": 20,
        "correct": 18,
        "accuracy": 0.9
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 70,
        "accuracy": 0.5428571428571428,
        "miss_count": 2,
        "false_alarm_count": 29
      },
      "SQ2": {
        "total": 38,
        "accuracy": 0.6052631578947368,
        "miss_count": 12,
        "false_alarm_count": 3
      },
      "SQ3": {
        "total": 71,
        "accuracy": 0.6338028169014085,
        "miss_count": 11,
        "false_alarm_count": 15
      },
      "SQ4": {
        "total": 59,
        "accuracy": 0.6101694915254238,
        "miss_count": 11,
        "false_alarm_count": 12
      },
      "SQ5": {
        "total": 62,
        "accuracy": 0.5967741935483871,
        "miss_count": 11,
        "false_alarm_count": 14
      }
    },
    "label": "weak_answers",
    "count": 300,
    "error_taxonomy": {
      "total_errors": 152,
      "error_distribution": {
        "MISS": 43,
        "PARSE_FAIL": 5,
        "WRONG_TYPE": 31,
        "FALSE_ALARM": 73
      },
      "error_by_sq": {
        "SQ1": {
          "PARSE_FAIL": 3,
          "FALSE_ALARM": 29
        },
        "SQ2": {
          "PARSE_FAIL": 1,
          "MISS": 11,
          "FALSE_ALARM": 3
        },
        "SQ3": {
          "MISS": 10,
          "WRONG_TYPE": 20,
          "PARSE_FAIL": 1,
          "FALSE_ALARM": 15
        },
        "SQ4": {
          "FALSE_ALARM": 12,
          "WRONG_TYPE": 11,
          "MISS": 11
        },
        "SQ5": {
          "MISS": 11,
          "FALSE_ALARM": 14
        }
      },
      "error_by_category": {
        "intrusion": 38,
        "elderly_specific": 15,
        "child_specific": 11,
        "behavioral_anomaly": 10
      }
    }
  }
}