autoformalization

LeanEuclid is a benchmark for autoformalization in the domain of Euclidean geometry, targeting the proof assistant Lean.

[ICML2026] The first, fully verified, sorry-free, large-scale Lean 4 library for statistical learning theory, covering infrastructures for mordern statistics and learning theory.

LeanInteract: A Python Interface for Lean 4

A graph-engineering agent harness for long-horizon, research-level autoformalization.

Workspace-first orchestration for long-horizon Lean 4 formalization agents.

DeonticBench

LLM-driven proof search on Erdős problems (2 solutions accepted on erdosproblems.com)

Autoformalization of Baek's solution to the moving sofa problem: the Gerver sofa is optimal

A calibrated faithfulness screen for informal↔Lean 4 pairs, on the command line and over MCP. Screens only, never certifies.

[ICLR 2026] Divide and Abstract: Autoformalization via Decomposition and Abstraction Learning

Formalization progress on the Ballantine-Beck-Feigon-Maurischat conjectures for subsum polynomials.

The AI mathematician that breaks conjectures before proving them, discharges every step inside a proof assistant, and hands you a proof you can check yourself.

Companion page for "Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery." A curated four-axis index covering MWPs, LLMs and reasoning models, multimodal geometry, Lean theorem proving, and verified discovery (FunSearch, AlphaEvolve, Erdős problems).

A selected list of Autoformalization papers, projects, frameworks, libraries, tools, and other resources.

Mathlib is the library of mathematics. This is the library of everything else that's provable - kernel-checked Lean 4 proofs of the mechanisms that run markets and institutions, formalized autonomously from plain-English blueprints.

Verifiable NL → CNL → KIF pipeline for military doctrine formalization. Research code for a 5-paper series on autoformalization in high-assurance domains.

Levee autoformalizes your rulebook (business logic, domain constraints, policy docs) into Lean 4 specs, then enforces them against AI agents at runtime. Formal verification for agentic AI: rules an LLM agent can’t talk its way around.

Official implementation of FormalEvolve, accepted to Findings of EMNLP 2026. Neuro-symbolic Lean 4 autoformalization.

Computational certificates for Graham/Alspach sequenceability across nine primes, dual-verified in two languages, with corrupted-certificate rejection tests