Reproduce “Specification-Guided Repair of Arithmetic Errors in Dafny Programs using LLMs” 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
Valentina Wu, Alexandra Mendes, Alexandre Abreu
The authors present a tool that repairs arithmetic bugs in Dafny by using contracts to identify suspicious code and prompting an LLM to propose replacement lines. Dafny verifies each patch against the specification. On artificially mutated DafnyBench programs, GPT-4o mini produced repairs that passed verification for 74.71% of ground_truth cases (1,985/2,657). This result assumes one bug per program and correct specifications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.