{
  "split": "train",
  "input_root": "data_dpo_full_log_v1",
  "total_pairs": 2500,
  "counters": {
    "chosen_rule_fallback": 1621,
    "rejected_weak_model": 1679,
    "chosen_strong_model": 879,
    "rejected_constructed_fallback": 821
  },
  "strong_answer_summary": {
    "total_episodes": 2500,
    "detection_accuracy": 0.5428,
    "miss_rate": 0.3529850746268657,
    "false_alarm_rate": 0.5577586206896552,
    "precision": 0.5726552179656539,
    "recall": 0.6470149253731343,
    "f1_security": 0.6075683251576735,
    "threat_type_accuracy": 0.5201845444059977,
    "parse_failure_rate": 0.0672,
    "breakdown": {
      "TP": {
        "total": 1340,
        "correct": 867,
        "accuracy": 0.6470149253731343
      },
      "FP": {
        "total": 980,
        "correct": 431,
        "accuracy": 0.43979591836734694
      },
      "TN": {
        "total": 180,
        "correct": 59,
        "accuracy": 0.3277777777777778
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 575,
        "accuracy": 0.5565217391304348,
        "miss_count": 26,
        "false_alarm_count": 226
      },
      "SQ2": {
        "total": 633,
        "accuracy": 0.6145339652448657,
        "miss_count": 196,
        "false_alarm_count": 45
      },
      "SQ3": {
        "total": 568,
        "accuracy": 0.47183098591549294,
        "miss_count": 166,
        "false_alarm_count": 123
      },
      "SQ4": {
        "total": 399,
        "accuracy": 0.5012531328320802,
        "miss_count": 60,
        "false_alarm_count": 135
      },
      "SQ5": {
        "total": 325,
        "accuracy": 0.5538461538461539,
        "miss_count": 25,
        "false_alarm_count": 118
      }
    },
    "label": "strong_answers",
    "count": 2500,
    "error_taxonomy": {
      "total_errors": 1559,
      "error_distribution": {
        "MISS": 402,
        "WRONG_TYPE": 416,
        "PARSE_FAIL": 100,
        "FALSE_ALARM": 641
      },
      "error_by_sq": {
        "SQ1": {
          "WRONG_TYPE": 65,
          "PARSE_FAIL": 29,
          "MISS": 3,
          "FALSE_ALARM": 223
        },
        "SQ2": {
          "MISS": 186,
          "WRONG_TYPE": 52,
          "PARSE_FAIL": 13,
          "FALSE_ALARM": 45
        },
        "SQ3": {
          "WRONG_TYPE": 136,
          "MISS": 145,
          "PARSE_FAIL": 33,
          "FALSE_ALARM": 122
        },
        "SQ4": {
          "WRONG_TYPE": 110,
          "MISS": 48,
          "PARSE_FAIL": 17,
          "FALSE_ALARM": 134
        },
        "SQ5": {
          "MISS": 20,
          "WRONG_TYPE": 53,
          "PARSE_FAIL": 8,
          "FALSE_ALARM": 117
        }
      },
      "error_by_category": {
        "intrusion": 331,
        "fire_gas": 107,
        "water_damage": 14,
        "device_fault": 68,
        "elderly_specific": 131,
        "child_specific": 84,
        "behavioral_anomaly": 83
      }
    }
  },
  "weak_answer_summary": {
    "total_episodes": 2500,
    "detection_accuracy": 0.528,
    "miss_rate": 0.2947761194029851,
    "false_alarm_rate": 0.6603448275862069,
    "precision": 0.5523085914669784,
    "recall": 0.7052238805970149,
    "f1_security": 0.6194690265486725,
    "threat_type_accuracy": 0.4804232804232804,
    "parse_failure_rate": 0.0508,
    "breakdown": {
      "TP": {
        "total": 1340,
        "correct": 945,
        "accuracy": 0.7052238805970149
      },
      "FP": {
        "total": 980,
        "correct": 320,
        "accuracy": 0.32653061224489793
      },
      "TN": {
        "total": 180,
        "correct": 55,
        "accuracy": 0.3055555555555556
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 575,
        "accuracy": 0.551304347826087,
        "miss_count": 7,
        "false_alarm_count": 250
      },
      "SQ2": {
        "total": 633,
        "accuracy": 0.5576619273301737,
        "miss_count": 148,
        "false_alarm_count": 127
      },
      "SQ3": {
        "total": 568,
        "accuracy": 0.46830985915492956,
        "miss_count": 139,
        "false_alarm_count": 154
      },
      "SQ4": {
        "total": 399,
        "accuracy": 0.47117794486215536,
        "miss_count": 68,
        "false_alarm_count": 140
      },
      "SQ5": {
        "total": 325,
        "accuracy": 0.6030769230769231,
        "miss_count": 33,
        "false_alarm_count": 95
      }
    },
    "label": "weak_answers",
    "count": 2500,
    "error_taxonomy": {
      "total_errors": 1671,
      "error_distribution": {
        "MISS": 334,
        "WRONG_TYPE": 491,
        "PARSE_FAIL": 89,
        "FALSE_ALARM": 757
      },
      "error_by_sq": {
        "SQ1": {
          "WRONG_TYPE": 45,
          "PARSE_FAIL": 9,
          "MISS": 1,
          "FALSE_ALARM": 248
        },
        "SQ2": {
          "MISS": 134,
          "WRONG_TYPE": 114,
          "PARSE_FAIL": 21,
          "FALSE_ALARM": 125
        },
        "SQ3": {
          "WRONG_TYPE": 163,
          "PARSE_FAIL": 31,
          "MISS": 120,
          "FALSE_ALARM": 151
        },
        "SQ4": {
          "WRONG_TYPE": 114,
          "PARSE_FAIL": 18,
          "MISS": 55,
          "FALSE_ALARM": 138
        },
        "SQ5": {
          "MISS": 24,
          "WRONG_TYPE": 55,
          "PARSE_FAIL": 10,
          "FALSE_ALARM": 95
        }
      },
      "error_by_category": {
        "intrusion": 336,
        "fire_gas": 126,
        "water_damage": 21,
        "device_fault": 46,
        "elderly_specific": 132,
        "child_specific": 86,
        "behavioral_anomaly": 78
      }
    }
  }
}