Reproduce “Whiel: A Neurosymbolic Platform for Certified Relational Program 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
Jesse Comer, Fangzhu Shen, Shuo Zhang, Mayur Naik, Sudeepa Roy, Val Tannen
Whiel checks whether relational programs satisfy correctness claims across all inputs, using an LLM to propose loop invariants or counterexamples that tools test; an optional Lean-based certifier checks proofs. On 86 curated tasks, GPT-5.5 Whiel’s verifier accepted answers for 73, versus 46 with one-round feedback and 20 with direct proof generation; acceptance does not mean every answer was certified.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.