metadata/split_spec.json

{
  "bucket_mix": "equal across training buckets; natural proportions within buckets",
  "confirm_subbucket_targets": {
    "r5_19": {
      "r12_15": 1000,
      "r16_19": 1000,
      "r5_7": 1000,
      "r8_11": 1000
    }
  },
  "counts": {
    "test_ood_confirm": 4000,
    "test_per_bucket": 1000,
    "train_per_bucket": 65536,
    "validation_per_bucket": 2048
  },
  "dataset_version": "1.1.0",
  "ood_bucket": "r5_19",
  "seed": "0",
  "selection": "metadata-stratified bottom-k SHA-256 rank",
  "test_subbucket_targets": {
    "r1_4": {
      "r1_2": 250,
      "r2_3": 250,
      "r3_4": 250,
      "r4_5": 250
    },
    "r5_19": {
      "r12_15": 250,
      "r16_19": 250,
      "r5_7": 250,
      "r8_11": 250
    }
  },
  "training_buckets": [
    "original",
    "r0",
    "r1_4"
  ]
}