manifest_cap8.json

{
  "source_dataset": "Zhongzhi1228/Recursive-Task-Synthesis-Trajectories",
  "trajectories_total": 327189,
  "eligible_after_gate": 60932,
  "eligible_groups": 1338,
  "per_group_cap": 8,
  "selected": 9479,
  "reconstructed": 9133,
  "after_dedup": 9080,
  "final_examples": 8886,
  "train_examples": 8686,
  "holdout_examples": 200,
  "max_seq_len": 32768,
  "token_stats": {
    "mean": 9481.661602520819,
    "p50": 8050.0,
    "p90": 16958.5,
    "p99": 28185.15,
    "max": 32329,
    "total_tokens": 84254045
  },
  "groups_covered": 1327,
  "model_mix": {
    "Qwen3.6-27B-base": 1383,
    "qwen35-27b-iter0000161-hf": 5608,
    "Qwen3.5-27B": 1421,
    "gpt-oss-120b": 474
  },
  "turns": {
    "mean": 11.996173756470853,
    "max": 60
  },
  "rewritten_turn_fraction": 0.574401020657048,
  "drop_counters": {
    "built": 9133,
    "dedup_command_signature": 53,
    "drop_missing_required_keys": 120,
    "drop_too_long": 194,
    "drop_unparseable": 226,
    "repaired_warning_preamble": 50169
  },
  "seed": 1228,
  "train_parquet": "data/sft-v1/rst_sft_train.parquet",
  "holdout_parquet": "data/sft-v1/rst_sft_holdout.parquet"
}