Metraipolis › archive › dc_9586b734eec2
Reap: Uncovering Monte Carlo proof search with a lightweight model — arXiv:2611.02678
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.02678
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T10:43:59.291Z
- SHA-256 of the archived text
7fe97bd87b87170511bb5b4a4dbd86de10c1be434bacc3e75a904344cb6f0ed5- Stable URL of this page
https://metraipolis.com/documento/dc_9586b734eec2
How to cite
Reap: Uncovering Monte Carlo proof search with a lightweight model — arXiv:2611.02678 — arXiv (en), https://arxiv.org/abs/2611.02678 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_9586b734eec2 (Metraipolis archive dc_9586b734eec2, sha256 7fe97bd87b87170511bb5b4a4dbd86de10c1be434bacc3e75a904344cb6f0ed5)
BibTeX
@misc{dc9586b734eec2,
title = {Reap: Uncovering Monte Carlo proof search with a lightweight model — arXiv:2611.02678},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.02678},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_9586b734eec2; sha256 7fe97bd87b87170511bb5b4a4dbd86de10c1be434bacc3e75a904344cb6f0ed5},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02678. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We present Reap, an open framework for learning and analyzing formal proof search in Lean4. Reap implements policy-value-guided Monte Carlo Tree Search directly in Lean, sharing proof-state transitions, AND-OR search, and replay verification between inference and training rollouts. Structured action logs expose how neural predictions and tactic execution interact during search. Using this framework, we train Reaper-1, a 1.7B-parameter policy-value model, through continued pre-training, supervised tactic learning, and iterative search-based reinforcement learning. Using a separate premise selector, the final system solves 79.9% of miniF2F-test at pass@32, establishing state-of-the-art performance among open provers of comparable size while remaining competitive with substantially larger models. Beyond improved solve rates, our results suggest that RL strengthens the prover's ability to organize proofs around useful intermediate results, favoring substantial advances over repeated local adjustments. This more directed proof construction is accompanied by increasingly elaborate tactic programs: fewer explicit search decisions are needed, while more work is performed within each decision. The resulting trade-off between search effort and tactic execution cost provides insight into how learning changes the prover's problem-solving strategy. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.02678 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-03T10:39:50.300Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).