# SCOPE-RL reproduction data

Prepared from the SCOPE-RL paper repo (github.com/tokencraft-lab/SCOPE-RL, arXiv 2607.11506) by `scope_rl/prep_data.py` in github.com/alphaXiv/paper-implementations.

- `dapo_math_2400.jsonl` - DAPO-Math 2400 training subset: chat prompt + rule-verifiable ground truth (sub-question decompositions stripped)
- `benchmark.jsonl` - paper eval benchmark: AIME24/25 (x8 rows), MATH500, GPQA-diamond