Can saved code-review rules catch a bug when it returns?
First-party technical note analyzing 428 saved Jev 1.13.0 public predictions for 214 before/after fix pairs. At score > 0.7, 107 broken methods and one repaired method are flagged. At score >= 0.7, 113 broken methods and one repaired method are flagged. Includes a pinned source revision and SHA-256, a runnable notebook, Python reproduction code, and CSV. No model requests are required to reproduce the calculations. Requirements were written with knowledge of the fixes; this is not a study of live agents or an identical-protocol ranking of PR review products. This is a first-party technical note by Perch / Verglas LLC. It is not a peer-reviewed paper. Source data: pinned public benchmark predictions. Source code: Perch on GitHub. Technical note and runnable notebook. Narrative and results are CC BY 4.0; reproduce.py and notebook code cells are MIT licensed. See LICENSE.txt for attribution and the complete code license.
Authors
- Verglas LLC
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-10-26
- DOI
- https://doi.org/10.5281/zenodo.23674954
- Primary Topic
- Software Engineering Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00