{
  "model": "Qwen/Qwen3.5-2B",
  "api_base": "http://localhost:8000/v1",
  "timestamp": "2026-05-10T13:55:09.665723",
  "total_evaluated": 1200,
  "avg_latency_seconds": 8.5,
  "api_errors": 0,
  "summary": {
    "total_episodes": 1200,
    "detection_accuracy": 0.6291666666666667,
    "miss_rate": 0.72,
    "false_alarm_rate": 0.05846153846153846,
    "precision": 0.8020833333333334,
    "recall": 0.28,
    "f1_security": 0.4150943396226416,
    "threat_type_accuracy": 0.551948051948052,
    "parse_failure_rate": 0.08833333333333333,
    "breakdown": {
      "TP": {
        "total": 550,
        "correct": 154,
        "accuracy": 0.28
      },
      "FP": {
        "total": 400,
        "correct": 365,
        "accuracy": 0.9125
      },
      "TN": {
        "total": 250,
        "correct": 236,
        "accuracy": 0.944
      }
    },
    "per_sq": {
      "SQ1": {
        "total": 190,
        "accuracy": 0.5578947368421052,
        "miss_count": 58,
        "false_alarm_count": 18
      },
      "SQ2": {
        "total": 240,
        "accuracy": 0.6208333333333333,
        "miss_count": 78,
        "false_alarm_count": 12
      },
      "SQ3": {
        "total": 290,
        "accuracy": 0.5551724137931034,
        "miss_count": 129,
        "false_alarm_count": 0
      },
      "SQ4": {
        "total": 290,
        "accuracy": 0.5689655172413793,
        "miss_count": 122,
        "false_alarm_count": 2
      },
      "SQ5": {
        "total": 190,
        "accuracy": 0.9157894736842105,
        "miss_count": 9,
        "false_alarm_count": 6
      }
    }
  },
  "errors": {
    "total_errors": 514,
    "error_distribution": {
      "PARSE_FAIL": 49,
      "FALSE_ALARM": 36,
      "MISS": 360,
      "WRONG_TYPE": 69
    },
    "error_by_sq": {
      "SQ1": {
        "PARSE_FAIL": 20,
        "FALSE_ALARM": 18,
        "MISS": 46,
        "WRONG_TYPE": 4
      },
      "SQ2": {
        "FALSE_ALARM": 12,
        "PARSE_FAIL": 6,
        "MISS": 73,
        "WRONG_TYPE": 4
      },
      "SQ3": {
        "MISS": 119,
        "PARSE_FAIL": 10,
        "WRONG_TYPE": 1
      },
      "SQ4": {
        "PARSE_FAIL": 7,
        "MISS": 118,
        "WRONG_TYPE": 2
      },
      "SQ5": {
        "FALSE_ALARM": 6,
        "PARSE_FAIL": 6,
        "WRONG_TYPE": 58,
        "MISS": 4
      }
    },
    "error_by_category": {
      "device_fault": 50,
      "fire_gas": 77,
      "intrusion": 135,
      "water_damage": 15,
      "behavioral_anomaly": 81,
      "child_specific": 28,
      "elderly_specific": 43
    }
  }
}