Submit a Proof
Submit a Lean 4 proof for any of the 300 benchmark conjectures. Human proofs are welcome. For AI model evaluation runs, contact us directly.
Proofs must be formalized in Lean 4
We only accept proofs written in Lean 4 against the formal-conjectures repository.
Proofs must compile with lake lean using Lean 4.27.0 and Mathlib. Natural language or informal proofs are not accepted.
All submissions are run through our zero-trust Comparator verification pipeline before appearing on the leaderboard. We will follow up by email with results.
lake lean with the formal-conjectures repo. Confirm your proof compiles before submitting.
Email Us Your Proof
Fill out the form below and click submit. This will open an email draft in your mail client with your proof attached. Nothing is stored on our servers until we verify it.
What happens after you submit?
We run your proof through our robust V2 zero-trust pipeline using Lean's official Comparator for sandboxed kernel replay, ensuring exact signature matching, axiom auditing (rejecting sorryAx), and 19 banned token scanning. This strictly prevents adversarial redefinition attacks. If your proof verifies, it appears on the leaderboard. We will email you with the results either way.
Verification Pipeline
sorry, admit, native_decide, unsafe, and 15 more. NFKC normalized to defeat homoglyph attacks.
Benchmark Download
lake lean examples and agent harness commands.
Model Evaluation
Want your AI model evaluated on the full benchmark? We run evaluations on our infrastructure. Contact us with your model name, organization, and intended scope.
Contact Us
Full Documentation
Benchmark format, API reference, and verification details.
Read docs →