{
  "model": "Qwen/Qwen3.5-9B",
  "api_base": "http://localhost:8000/v1",
  "timestamp": "2026-05-04T15:54:43.664262",
  "total_evaluated": 60,
  "summary": {
    "total_episodes": 60,
    "detection_accuracy": 0.55,
    "miss_rate": 0.0,
    "false_alarm_rate": 0.8125,
    "precision": 0.5185185185185185,
    "recall": 1.0,
    "f1_security": 0.6829268292682926,
    "threat_type_accuracy": 0.2857142857142857,
    "parse_failure_rate": 0.016666666666666666,
    "breakdown": {
      "TP": {
        "total": 28,
        "correct": 28,
        "accuracy": 1.0
      },
      "FP": {
        "total": 28,
        "correct": 3,
        "accuracy": 0.10714285714285714
      },
      "TN": {
        "total": 4,
        "correct": 2,
        "accuracy": 0.5
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 13,
        "accuracy": 0.6923076923076923,
        "miss_count": 0,
        "false_alarm_count": 4
      },
      "SQ2": {
        "total": 6,
        "accuracy": 0.5,
        "miss_count": 0,
        "false_alarm_count": 3
      },
      "SQ3": {
        "total": 17,
        "accuracy": 0.5882352941176471,
        "miss_count": 0,
        "false_alarm_count": 7
      },
      "SQ4": {
        "total": 17,
        "accuracy": 0.47058823529411764,
        "miss_count": 0,
        "false_alarm_count": 9
      },
      "SQ5": {
        "total": 7,
        "accuracy": 0.42857142857142855,
        "miss_count": 0,
        "false_alarm_count": 3
      }
    }
  },
  "errors": {
    "total_errors": 47,
    "error_distribution": {
      "FALSE_ALARM": 26,
      "WRONG_TYPE": 20,
      "PARSE_FAIL": 1
    },
    "error_by_sq": {
      "SQ1": {
        "FALSE_ALARM": 4
      },
      "SQ2": {
        "FALSE_ALARM": 3,
        "WRONG_TYPE": 2
      },
      "SQ3": {
        "FALSE_ALARM": 7,
        "WRONG_TYPE": 7
      },
      "SQ4": {
        "FALSE_ALARM": 9,
        "WRONG_TYPE": 8
      },
      "SQ5": {
        "FALSE_ALARM": 3,
        "PARSE_FAIL": 1,
        "WRONG_TYPE": 3
      }
    },
    "error_by_category": {
      "water_damage": 2,
      "intrusion": 4,
      "behavioral_anomaly": 4,
      "child_specific": 4,
      "elderly_specific": 3,
      "fire_gas": 3
    }
  },
  "pipeline": "EGPv2"
}