{
  "comparison_unit": "five paired repeated calls on the same fixed case set; not five independent studies",
  "dataset": "defect-lab-micro-v1",
  "evidence_label": "local_reconstruction_experiment",
  "limits": [
    "source-derived simplified reconstructions",
    "only four test benchmark families",
    "counterfactual repair controls",
    "uncontrolled provider sampling",
    "pretraining contamination unknown",
    "no TVB parity or superiority claim"
  ],
  "model": "gpt-5.5",
  "runs": [
    {
      "condition": "baseline",
      "metrics": {
        "clean_accuracy": 0.8571428571428571,
        "f1": 0.9333333333333333,
        "false_positive_rate": 0.14285714285714285,
        "fn": 0,
        "fp": 1,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 0.875,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 1
    },
    {
      "condition": "fewshot",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 1
    },
    {
      "condition": "baseline",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 2
    },
    {
      "condition": "fewshot",
      "metrics": {
        "clean_accuracy": 0.8571428571428571,
        "f1": 0.9333333333333333,
        "false_positive_rate": 0.14285714285714285,
        "fn": 0,
        "fp": 1,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 0.875,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 2
    },
    {
      "condition": "baseline",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 3
    },
    {
      "condition": "fewshot",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 3
    },
    {
      "condition": "baseline",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 4
    },
    {
      "condition": "fewshot",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 4
    },
    {
      "condition": "baseline",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 5
    },
    {
      "condition": "fewshot",
      "metrics": {
        "clean_accuracy": 1.0,
        "f1": 1.0,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 1.0,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      },
      "repeat": 5
    }
  ],
  "schema": "defectlab.public-analysis.v1",
  "summary": {
    "baseline": {
      "completed_repeats": 5,
      "mean": {
        "clean_accuracy": 0.9714285714285714,
        "f1": 0.9866666666666667,
        "false_positive_rate": 0.02857142857142857,
        "fn": 0,
        "fp": 0.2,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 0.975,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      }
    },
    "fewshot": {
      "completed_repeats": 5,
      "mean": {
        "clean_accuracy": 0.9714285714285714,
        "f1": 0.9866666666666667,
        "false_positive_rate": 0.02857142857142857,
        "fn": 0,
        "fp": 0.2,
        "known_abstentions": 0,
        "localized_tp": 7,
        "n_clean": 7,
        "n_positive": 7,
        "precision": 0.975,
        "recall": 1.0,
        "tp": 7,
        "unknown_abstentions": 2,
        "unknown_defect_claims": 0,
        "witness_reproduced_tp": 7
      }
    }
  }
}