一句话：针对知识流失风险高的 PR 提出增加额外评审者的推荐器，量化其降低流失风险与增加评审工作量之间的权衡。
开发者流失不可避免，会造成知识丢失、生产力下降与缺陷增加。已有工作只在“替换”层面量化评审者推荐器的知识分布能力，未考虑增加额外评审者。本文聚焦包含被废弃或被囤积文件的 pull request——这类 PR 更易因开发者流失造成严重知识损失，称为“知识因流失而处于风险的 PR”（PRKR）。作者提出为 PRKR 推荐额外评审者以更广泛地分散知识，同时限制团队额外工作量。最激进的推荐器 TurnoverRec++ 对所有 PRKR 都增加一名额外评审者，平均可降低 82% 的流失风险（-82%），但代价显著：平均 16.6% 的 pull request 会被加入额外评审者。模拟显示降低风险文件数与额外评审量之间存在紧耦合的权衡，且该权衡是非线性的：用相对较少的额外评审即可显著降低流失风险，例如新推荐器 FarAwareRec 降低 42% 的风险文件，却只对 1.9% 的 PR 增加额外评审者。作者认为管理者应结合流失率与可用预算等因素选择策略，并提供复现包。
Developer turnover is inevitable in software projects, leading to knowledge loss, reduced productivity, and increased defects. Prior works have quantified the knowledge distribution ability of reviewer recommenders at the replacement level. However, they did not consider adding reviewers. In this work, we target pull requests containing abandoned or hoarded files, which are more prone to significant knowledge loss from developer turnover, i.e. Pull Requests with Knowledge at Risk to Turnover (PRKRs). On PRKRs, we know that the knowledge is not distributed across the team. In this study, we aim to suggest an extra reviewer for PRKRs to spread their knowledge more broadly among the developers while limiting the additional work for the development team. Our most aggressive recommender is TurnoverRec++, which adds an extra reviewer to all PRKRs. This recommender achieves an average reduction of -82% in turnover risk, however, there is a significant cost, with extra reviewers being added to an average 16.6% of all pull requests. Our simulations show a tight trade-off between reducing the number of files at risk to turnover and the extra reviews. We also show that this trade-off is non-linear, and it is possible to significantly decrease the turnover risk with relatively few extra reviews. For example, the novel recommender, FarAwareRec, reduces files at risk to turnover by -42% but only adds an extra reviewer to 1.9%. Ultimately, our results show that managers should determine which strategy best suits their situation by considering factors such as turnover rates and available budgets. Our scripts and data are available in our replication package [17].
@inproceedings{paperbot11621,
title = {The Cost vs the Benefit of Adding an Extra Code Reviewer to Mitigate Developer Turnover through Reviewer Recommenders},
author = {Mohammadali Sefidi Esfahani and Fahimeh Hajari and Peter C. Rigby},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3744916.3764545}
}