{
  "split": "dev",
  "input_root": "data_dpo_full_log_v1",
  "total_pairs": 300,
  "counters": {
    "chosen_rule_fallback": 201,
    "rejected_weak_model": 190,
    "rejected_constructed_fallback": 110,
    "chosen_strong_model": 99
  },
  "strong_answer_summary": {
    "total_episodes": 300,
    "detection_accuracy": 0.5433333333333333,
    "miss_rate": 0.25625,
    "false_alarm_rate": 0.6571428571428571,
    "precision": 0.5639810426540285,
    "recall": 0.74375,
    "f1_security": 0.6415094339622642,
    "threat_type_accuracy": 0.5714285714285714,
    "parse_failure_rate": 0.11333333333333333,
    "breakdown": {
      "TP": {
        "total": 160,
        "correct": 119,
        "accuracy": 0.74375
      },
      "FP": {
        "total": 120,
        "correct": 35,
        "accuracy": 0.2916666666666667
      },
      "TN": {
        "total": 20,
        "correct": 9,
        "accuracy": 0.45
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 70,
        "accuracy": 0.5428571428571428,
        "miss_count": 5,
        "false_alarm_count": 27
      },
      "SQ2": {
        "total": 38,
        "accuracy": 0.631578947368421,
        "miss_count": 10,
        "false_alarm_count": 4
      },
      "SQ3": {
        "total": 71,
        "accuracy": 0.4507042253521127,
        "miss_count": 14,
        "false_alarm_count": 25
      },
      "SQ4": {
        "total": 59,
        "accuracy": 0.6610169491525424,
        "miss_count": 5,
        "false_alarm_count": 14
      },
      "SQ5": {
        "total": 62,
        "accuracy": 0.4838709677419355,
        "miss_count": 7,
        "false_alarm_count": 22
      }
    },
    "label": "strong_answers",
    "count": 300,
    "error_taxonomy": {
      "total_errors": 188,
      "error_distribution": {
        "MISS": 25,
        "PARSE_FAIL": 21,
        "WRONG_TYPE": 51,
        "FALSE_ALARM": 91
      },
      "error_by_sq": {
        "SQ1": {
          "WRONG_TYPE": 6,
          "PARSE_FAIL": 6,
          "FALSE_ALARM": 26
        },
        "SQ2": {
          "MISS": 10,
          "WRONG_TYPE": 3,
          "FALSE_ALARM": 4
        },
        "SQ3": {
          "WRONG_TYPE": 19,
          "MISS": 7,
          "PARSE_FAIL": 7,
          "FALSE_ALARM": 25
        },
        "SQ4": {
          "FALSE_ALARM": 14,
          "PARSE_FAIL": 3,
          "WRONG_TYPE": 17,
          "MISS": 3
        },
        "SQ5": {
          "MISS": 5,
          "PARSE_FAIL": 5,
          "WRONG_TYPE": 6,
          "FALSE_ALARM": 22
        }
      },
      "error_by_category": {
        "intrusion": 38,
        "device_fault": 6,
        "elderly_specific": 12,
        "child_specific": 11,
        "behavioral_anomaly": 9
      }
    }
  },
  "weak_answer_summary": {
    "total_episodes": 300,
    "detection_accuracy": 0.5566666666666666,
    "miss_rate": 0.2,
    "false_alarm_rate": 0.6928571428571428,
    "precision": 0.5688888888888889,
    "recall": 0.8,
    "f1_security": 0.664935064935065,
    "threat_type_accuracy": 0.5546875,
    "parse_failure_rate": 0.07666666666666666,
    "breakdown": {
      "TP": {
        "total": 160,
        "correct": 128,
        "accuracy": 0.8
      },
      "FP": {
        "total": 120,
        "correct": 33,
        "accuracy": 0.275
      },
      "TN": {
        "total": 20,
        "correct": 6,
        "accuracy": 0.3
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 70,
        "accuracy": 0.5714285714285714,
        "miss_count": 0,
        "false_alarm_count": 30
      },
      "SQ2": {
        "total": 38,
        "accuracy": 0.631578947368421,
        "miss_count": 8,
        "false_alarm_count": 6
      },
      "SQ3": {
        "total": 71,
        "accuracy": 0.4647887323943662,
        "miss_count": 11,
        "false_alarm_count": 27
      },
      "SQ4": {
        "total": 59,
        "accuracy": 0.5423728813559322,
        "miss_count": 7,
        "false_alarm_count": 20
      },
      "SQ5": {
        "total": 62,
        "accuracy": 0.6129032258064516,
        "miss_count": 6,
        "false_alarm_count": 14
      }
    },
    "label": "weak_answers",
    "count": 300,
    "error_taxonomy": {
      "total_errors": 190,
      "error_distribution": {
        "MISS": 21,
        "WRONG_TYPE": 57,
        "PARSE_FAIL": 15,
        "FALSE_ALARM": 97
      },
      "error_by_sq": {
        "SQ1": {
          "WRONG_TYPE": 5,
          "FALSE_ALARM": 30
        },
        "SQ2": {
          "WRONG_TYPE": 8,
          "MISS": 7,
          "PARSE_FAIL": 1,
          "FALSE_ALARM": 6
        },
        "SQ3": {
          "MISS": 6,
          "WRONG_TYPE": 22,
          "PARSE_FAIL": 5,
          "FALSE_ALARM": 27
        },
        "SQ4": {
          "FALSE_ALARM": 20,
          "WRONG_TYPE": 15,
          "MISS": 5,
          "PARSE_FAIL": 2
        },
        "SQ5": {
          "MISS": 3,
          "PARSE_FAIL": 7,
          "WRONG_TYPE": 7,
          "FALSE_ALARM": 14
        }
      },
      "error_by_category": {
        "intrusion": 38,
        "water_damage": 3,
        "device_fault": 5,
        "elderly_specific": 15,
        "child_specific": 9,
        "behavioral_anomaly": 8
      }
    }
  }
}