Solved lean-eval benchmark workspaces, one per subdirectory.
This repository is a submission source, not the submission target. To score
these on the public leaderboard, open a
"Submit benchmark solution" issue on
leanprover/lean-eval-submissions
and point its Submission URL at this repo. That CI walks every subdirectory,
finds each lakefile.toml whose name matches a benchmark problem id with a
Submission.lean alongside, and scores it with the comparator. Only
Submission.lean and files under Submission/ are ever read.
| Problem | Status | Model |
|---|---|---|
baer_suzuki |
✅ comparator-accepted (local lake test) |
Claude Opus 4.7 + GPT-5.5 (human-in-the-loop) |
Each <problem>/ is a self-contained comparator workspace generated from
leanprover/lean-eval:
Submission.lean,Submission/— the solution (solver-owned)Challenge.lean,Solution.lean,ChallengeDeps.lean,config.json,lakefile.toml, … — trusted scaffold, identical to the benchmark'sgenerated/<problem>/
cd <problem>
lake testRequires the comparator toolchain (comparator, landrun, lean4export) on
PATH; see the lean-eval README.