AI

When Correctness Feedback Erases Evidence: Phase-Separated Completeness Auditing in Verus

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 16/10/2026, 04:03:24 UTC.

Briefings

A new study on Verus/Rust tasks demonstrates that correctness feedback suppresses evidence needed for specification completeness auditing. Independent discovery yields 59.0% certified witnesses, while feedback-only methods achieve only 27.1%. Phase-separated execution matches aggregate yield without reusing responses, and preserving witnesses improves specification repair from 62.5% to 81.3%. The effect is specific to Verus/Rust and absent in Dafny/Lean evaluations.

Key points

• Correctness feedback in Verus/Rust tasks suppresses evidence needed for specification completeness auditing, reducing certified witness yield from 59.0% to 27.1%.
• Phase-separated execution matches aggregate witness yield without reusing responses, while preserving witnesses improves specification repair from 62.5% to 81.3%.

Context

This research investigates the trade-off between correctness feedback and specification completeness auditing in Verus/Rust tasks, showing that feedback can erase evidence needed to verify contracts.

Limitations

The study covers only 12 Verus/Rust tasks across four model families; effects are absent in Dafny and Lean evaluations, and the specific impact of parser repair on yield remains unquantified.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 16/10/2026, 05:02:17 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeifdiv3ficczamooyr7q5clswrbdzilv3sa5ahapu6kzr6o7fnv4na