Zihao Zhang and co-authors submitted VeriBase on 1 November 2026, proposing a framework that automatically builds reusable software-verification libraries from successful verification solutions. The arXiv abstract says the framework generalizes implementations together with their dependent theorem statements and proofs, addressing work that otherwise requires substantial expert effort.[1]

VeriBase groups code and its associated theorems into “declaration families.” Its three phases prove seed tasks, generalize code and theorems, and harvest the resulting families into shared interfaces. The authors say it updates dependent declarations when interfaces change and uses Lean to check the resulting code and proofs. This makes the relationship between changed implementations and dependent proofs central to the proposal.[1]

The authors report evaluation across four benchmarks with three backend large language models (LLMs). They say a general-purpose library built from 4,191 independent programs increases average task success rate by up to 17.0% and reduces proof completion time by up to 21.4%. The abstract also reports further improvement with benchmark-specific libraries, without quantifying it. These are author-reported results; the saved abstract does not specify whether the success-rate increase is relative or in percentage points.[1]

[1]: https://arxiv.org/abs/2611.01743

Sources

  1. VeriBase: Automated Construction of Theorem Libraries for Software VerificationarXiv · 1 Nov 2026

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.

e3a1be79d14a885100726471c55556def95afaddbe39dc58bba7fd0954fcdb29