experiments/trained.json

{
  "agent": "trained",
  "model": "gauri0508/med-record-audit-qwen2.5-3b-grpo",
  "n_trials_per_case": 10,
  "average_score": 0.3546,
  "peak_average_score": 0.6351,
  "per_case": {
    "easy_001": {
      "difficulty": "easy",
      "all_scores": [
        0.8167,
        0.8167,
        0.81,
        0.8167,
        0.8167,
        0.63,
        0.8167,
        0.8167,
        0.8167,
        0.63
      ],
      "avg_score": 0.7787,
      "best_score": 0.8167,
      "best_rubric": {
        "finding_accuracy": 0.25,
        "evidence_validity": 0.2,
        "completeness": 0.2,
        "efficiency": 0.0667,
        "anti_hacking": 0.1
      },
      "n_trials": 10
    },
    "medium_001": {
      "difficulty": "medium",
      "all_scores": [
        0.34,
        0.3827,
        0.01,
        0.01,
        0.324,
        0.01,
        0.01,
        0.2178,
        0.01,
        0.6853
      ],
      "avg_score": 0.2,
      "best_score": 0.6853,
      "best_rubric": {
        "finding_accuracy": 0.2,
        "evidence_validity": 0.2,
        "completeness": 0.1333,
        "efficiency": 0.052,
        "anti_hacking": 0.1
      },
      "n_trials": 10
    },
    "hard_001": {
      "difficulty": "hard",
      "all_scores": [
        0.01,
        0.4033,
        0.01,
        0.01,
        0.01,
        0.01,
        0.01,
        0.3666,
        0.01,
        0.01
      ],
      "avg_score": 0.085,
      "best_score": 0.4033,
      "best_rubric": {
        "finding_accuracy": 0.15,
        "evidence_validity": 0.0667,
        "completeness": 0.0333,
        "efficiency": 0.0533,
        "anti_hacking": 0.1
      },
      "n_trials": 10
    }
  }
}