Review recall and precision

Last updated 2 October 2026.

What are recall and precision in code review?

Recall is the share of real defects a review finds, and precision is the share of its findings that are real; a useful reviewer needs both, since high recall with low precision buries the real bugs in noise.

How to measure them

• Take real pull requests with known, hand-labelled defects.
• Recall = labelled defects found ÷ labelled defects.
• Precision = real findings ÷ all findings, where each unmatched finding is checked against the code.
• Review each pull request more than once: results vary between runs.

MergeSafe’s numbers

• Strong: 78% recall (47 of 60) and 98% precision (172 of 175).
• Budget: 72% recall (43 of 60) and 99% precision (122 of 123).
• On 11 real pull requests, two reviews each, September 2026. A small set: a guide, not a guarantee.

The method: Benchmarks. Related: false positive.

Next: AI code review →