_work/final_representative_counts.json

{
  "representatives_only": true,
  "source": "data_v3_rebuild/train|val|test parquet outputs",
  "split_rows": {
    "test": 331455,
    "train": 36888301,
    "val": 373637
  },
  "split_unique_seq_ids": {
    "test": 331455,
    "train": 36888301,
    "val": 373637
  }
}