{
  "inputs": [
    "scripts/clinical-dataset/medstack-bedside-india-v2.jsonl",
    "scripts/clinical-dataset/medstack-bedside-row-v2.jsonl"
  ],
  "prefix": "scripts/clinical-dataset/eval5-splits/bedside",
  "seed": 42,
  "val_pct": 3,
  "test_pct": 2,
  "dedup": true,
  "struct_cap": 300,
  "generated_at": "2026-07-06T11:15:21.411Z",
  "before": {
    "rows_scanned": 1260000
  },
  "removed": {
    "invalid_json": 0,
    "empty_required_field": 0,
    "exact_duplicates": 0,
    "over_cap_skeletons": 0
  },
  "benchmark_guard": {
    "train_only_match": [],
    "rows_forced_to_train": 0,
    "skeleton_conflicts": 0
  },
  "after": {
    "rows_kept": 1260000,
    "unique_skeletons": 1259481,
    "new_exact_dup_pct": 0,
    "new_structural_dup_pct": 0.04,
    "splits": {
      "train": 1196848,
      "val": 37879,
      "test": 25273
    },
    "split_pct": {
      "train": 94.99,
      "val": 3.01,
      "test": 2.01
    }
  },
  "leakage_guarantee": "split bucket = hash(structural output key); every exact/structural near-dup routes to one split. val/test skeletons disjoint from train by construction.",
  "per_domain_coverage": {
    "bedside_specialties": {
      "train": 1196848,
      "val": 37879,
      "test": 25273
    }
  }
}