CITATION.cff

cff-version: 1.2.0
message: "If you use this benchmark, please cite it as below."
title: "SWE-Review-Bench"
type: software
authors:
  - family-names: Zhou
    given-names: Zhuangyu
    email: lmnstzz@gmail.com
    affiliation: "University of Bonn"
repository-code: "https://github.com/lmnst/SWE-Review-Bench"
url: "https://github.com/lmnst/SWE-Review-Bench"
abstract: "SWE-Review-Bench is a pilot benchmark for cold code-review bug finding on SWE-bench Lite. It evaluates LLM-based and static reviewers under a no-leakage input policy: the reviewer sees only the pre-fix file content, and the oracle (the fix patch's line ranges) is read solely by the scorer. The v0.1 release covers a 20-instance pilot with three reviewers, a prompt-variant probe, and a pytest leakage suite."
keywords:
  - code-review
  - benchmark
  - large-language-models
  - software-engineering
  - SWE-bench
license: MIT
version: "0.1-pilot"
date-released: 2026-05-12