Reproduce “Formal Disco: Scalable Open-Ended Generation of Formally Verified Programs” 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
Gabriel Poesia, Simon Henniger, Tzu-Han Hsu, Yilun Du, Nada Amin
Formal Disco generates formally verified programs across Dafny, Verus, and Frama-C by having language-model workers draft, repair, and extend code using compiler and verifier feedback, then selecting successful examples for feature diversity. Fine-tuning Qwen 2.5-Coder 32B on its data reaches 43.0% Pass@1 on VerusBench, matching Claude Opus 4.5’s single-attempt score; this result is limited to that task setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.