Reproduce “From Natural Language to Verified Code: Toward AI Assisted Problem-to-Code Generation with Dafny-Based Formal Verification” 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
Md Erfan, Md Kamal Hossain Chowdhury, Ahmed Ryan, Md Rayhanur Rahman
The authors evaluate open-weight models generating Dafny code with formal contracts and loop annotations. They compare natural-language prompts, method signatures, and iterative verifier feedback on 11 UVa-derived problems. At T=0.2, GPT-OSS-120B improves from 0/11 contextless healing to 9/11 with signature-guided healing. These are verifier successes, not functional-test passes, and apply only to the evaluated tasks, models, and Dafny setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.