LeanGraphSearch mines Mathlib dependency graphs, lifting proof success to 81%
kaiwei_chang · x · 2026-10-17
LeanGraphSearch starts from a key insight: the lemma an AI needs to prove a theorem may look nothing like the theorem itself, so similarity-based retrieval misses crucial ingredients.
- Instead of query matching, it follows formal dependency edges behind related Mathlib results to uncover missing proof ingredients multi-hop
- Results: 94.7% recall on MathlibQR (+12.0pp vs LeanSearch v2), 42.8% coverage on MathlibMPR (+3.7pp), and 81% success on FATE-H vs 73% for LeanSearch v2 and 52% with no retrieval — yielding more Lean-verified proofs
More from coding & agent
- A one-shot cron job prompt got Claude to crack ciphers and advance a 1932 number theory bound — IgorCarron · 2026-10-17
- Backend migration from Python to Go now takes one engineer just one week — rakyll · 2026-10-17
- Audio founder says Claude one-shot decompiled rival algorithms, moat gone — ChrisGPT · 2026-10-17
- Google's playbook for agent evals: four layers of checks beyond the vibe check — rseroter · 2026-10-17
- AgentForesight: a 7B online auditor that catches multi-agent failures mid-run — jiqizhixin · 2026-10-17
- Qwen3.8-27B shows a weird write-delete loop in /tmp before every web research task — Training-Respect8066 · 2026-10-17