{
  "method": "Paired descriptive reanalysis of already published fixed predictions. No new model or product run. No population significance claim.",
  "sample_n": 300,
  "source_sha256": "a7b33e6e771eba7b040f351e87cca909155e63aca8e0290a79f2d7fa3e850e28",
  "python": "3.14.8",
  "comparison": {
    "before": "bm25_or_none",
    "after": "bm25_name_description"
  },
  "paired_correctness": {
    "both_correct": 96,
    "both_wrong": 186,
    "correction": 15,
    "regression": 3
  },
  "errors": {
    "bm25_or_none": {
      "correct": 99,
      "missed_tool": 49,
      "unnecessary_tool": 34,
      "wrong_tool": 118
    },
    "bm25_name_description": {
      "correct": 111,
      "missed_tool": 46,
      "unnecessary_tool": 36,
      "wrong_tool": 107
    }
  },
  "none_gold_n": 66,
  "tool_gold_n": 234
}