Research · Updated 16 Oct, 09:30 am IST
Correctness feedback can erase evidence in specification-completeness auditing
Why it matters for readers: It matters because some ways of giving feedback during program checking can accidentally hide the very problems audits are meant to reveal.
- Correctness feedback reduces the yield of certified witnesses compared with independent no-feedback discovery (59.0% vs 27.1%).1
- A format-matched iterative control attributes most of the gap to feedback content rather than to independent restarts.1
- A phase-separated execution matches the higher aggregate yield without reusing responses, and most certified witnesses also fail held-out cases.1
- Preserving admitted failures before repair produces stronger specification repair: Phase-Revised contracts reach 81.3% implementation success versus 62.5% for IntentShield (IS)-Revised.1
- The negative effect of corrective feedback did not reproduce in available Dafny and Lean evaluations, delimiting the empirical finding to the studied Verus/Rust workflow.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started