🧪 Test?View on arXiv
When Correctness Feedback Erases Evidence: Phase-Separated Completeness Auditing in Verus
Author1, Author2, Author3, Author4, Author5
correctness auditingspecification repairprogram verification
2610.20676
Builder Relevance
1h ago80%
Abstract
The paper discusses how correctness feedback can obscure evidence in specification-completeness auditing, demonstrating that independent no-feedback discovery yields significantly better results.
Reality Card
Core Claim
Independent no-feedback discovery more than doubles the certified-witness yield compared to IntentShield's corrective-feedback loop, achieving 59.0% versus 27.1%.
Method / Result
Phase-Revised contracts achieve 81.3% implementation success compared to 62.5% for IS-Revised.
Limitations
The effect does not reproduce in available Dafny and Lean evaluations, indicating a limitation in generalizability.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.