Reproduce “Automating Formal Verification with Agent-Guided Tree Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Leo Yao
The authors build a vericoding agent that generates code and a formal proof for a specification, searches Lean’s mathlib for lemmas, and uses Lean’s feedback to guide further attempts. With GPT-5.4, the agent solves 95.0% of 423 benchmark specifications using up to 50 model calls per specification. This is a benchmark result, not evidence of production readiness or arbitrary-task performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.