AI, Data Science & Cloud Certificates from Google, IBM & Meta — 40% Off
One plan covers every Professional Certificate on Coursera. 40% off Coursera Plus Annual.
Unlock All Certificates
This conference talk explores the challenges of proof instability in SMT-based program verification systems like Dafny and F*. Discover how researchers Yi Zhou and Bryan Parno from Carnegie Mellon University address situations where minor changes to source code can cause verification failures. Learn about their methodology for detecting instability, findings from measurement studies across five verification query sets, and their pruning algorithm designed to reduce instability by eliminating irrelevant SMT query context. The presentation also covers ongoing research aimed at controlling relevant but unstable context, with the ultimate goal of achieving stable proofs while maintaining powerful automation capabilities. This 15-minute talk was presented at the Dafny 2025 workshop on January 19, 2025, sponsored by ACM SIGPLAN.
