Qodo/PR-Review-Bench

Qodo代码审查基准1.0是一个大规模评估数据集，旨在测量AI驱动的代码审查系统在实际拉取请求场景中的有效性。数据集包含100个真实的、已合并的拉取请求，这些请求来自多个编程语言（TypeScript、Python、JavaScript、C、C#、Rust和Swift）的生产级开源仓库，并注入了580个精心设计的问题。这些问题包括功能性错误和最佳实践违规，从而能够同时评估代码的正确性和质量。数据集由Qodo团队精心策划，并通过双重验证过程确保每个修改后的拉取请求的准确性。

The Qodo Code Review Benchmark 1.0 is a large-scale evaluation dataset designed to measure the effectiveness of AI-powered code review systems in realistic pull request scenarios. The dataset consists of 100 real, merged pull requests sourced from production-grade open-source repositories across multiple languages (TypeScript, Python, JavaScript, C, C#, Rust, and Swift), into which 580 carefully injected issues were introduced. These issues include both functional bugs and best-practice violations, enabling simultaneous evaluation of code correctness and code quality. The dataset was curated by the Qodo team and underwent a double validation process to ensure the accuracy of each modified pull request.