Pedro Ribeiro, Martim Afonso, André Vasconcelos and João F. Ferreira introduced VerusWeave in an arXiv submission dated 30 October 2026. The authors describe a benchmark for coding agents repairing mathematical proofs in concurrent Verus programs, addressing limited concurrency coverage in existing Verus benchmarks.[1]
Verus extends Rust with formal verification capabilities. According to the abstract, VerusWeave draws on four open-source projects and contains 75 focused tasks and 17 full-module variants. Tasks remove selected proof annotations; agents must restore the missing material while preserving executable code and specifications. This constraint makes preservation of the original verification target part of the evaluation.[1]
The authors say their evaluation harness checks repairs with Verus and a change checker, recording solutions, execution traces and resource usage. Their initial evaluation of four general-purpose coding-agent harnesses reports success rates of 75–85% on focused tasks and 29–47% on full-module variants. These are author-reported results; the saved abstract does not identify the four harnesses or provide their configurations or detailed scoring rules.[1]
[1]: https://arxiv.org/abs/2611.00287
Sources
Check or correct this page
Verification record
The digest below shows whether the text you are reading matches the published source. It does not show that the reporting is correct. The build computes it from this page's Markdown source and lists it in the site manifest.
186db4e86f3fd27ef96b85520f237410e878c824e141c3a957697c6286bc9bb6