{
  "model": "Qwen/Qwen2.5-7B-Instruct",
  "api_base": "http://localhost:8000/v1",
  "timestamp": "2026-04-30T22:18:00.620006",
  "total_evaluated": 1200,
  "avg_latency_seconds": 7.61,
  "api_errors": 0,
  "summary": {
    "total_episodes": 1200,
    "detection_accuracy": 0.5975,
    "miss_rate": 0.8781818181818182,
    "false_alarm_rate": 0.0,
    "precision": 1.0,
    "recall": 0.12181818181818181,
    "f1_security": 0.2171799027552674,
    "threat_type_accuracy": 0.44776119402985076,
    "parse_failure_rate": 0.04,
    "breakdown": {
      "TP": {
        "total": 550,
        "correct": 67,
        "accuracy": 0.12181818181818181
      },
      "FP": {
        "total": 400,
        "correct": 400,
        "accuracy": 1.0
      },
      "TN": {
        "total": 250,
        "correct": 250,
        "accuracy": 1.0
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 190,
        "accuracy": 0.5894736842105263,
        "miss_count": 78,
        "false_alarm_count": 0
      },
      "SQ2": {
        "total": 240,
        "accuracy": 0.5791666666666667,
        "miss_count": 101,
        "false_alarm_count": 0
      },
      "SQ3": {
        "total": 290,
        "accuracy": 0.5551724137931034,
        "miss_count": 129,
        "false_alarm_count": 0
      },
      "SQ4": {
        "total": 290,
        "accuracy": 0.5517241379310345,
        "miss_count": 130,
        "false_alarm_count": 0
      },
      "SQ5": {
        "total": 190,
        "accuracy": 0.7631578947368421,
        "miss_count": 45,
        "false_alarm_count": 0
      }
    }
  },
  "errors": {
    "total_errors": 520,
    "error_distribution": {
      "MISS": 460,
      "WRONG_TYPE": 37,
      "PARSE_FAIL": 23
    },
    "error_by_sq": {
      "SQ1": {
        "MISS": 76,
        "WRONG_TYPE": 2,
        "PARSE_FAIL": 2
      },
      "SQ2": {
        "MISS": 98,
        "PARSE_FAIL": 3,
        "WRONG_TYPE": 1
      },
      "SQ3": {
        "MISS": 126,
        "PARSE_FAIL": 3,
        "WRONG_TYPE": 1
      },
      "SQ4": {
        "MISS": 125,
        "PARSE_FAIL": 5
      },
      "SQ5": {
        "PARSE_FAIL": 10,
        "MISS": 35,
        "WRONG_TYPE": 33
      }
    },
    "error_by_category": {
      "device_fault": 138,
      "fire_gas": 116,
      "water_damage": 52,
      "intrusion": 115,
      "behavioral_anomaly": 35,
      "child_specific": 14,
      "elderly_specific": 27
    }
  }
}