Review precision
Review precision is the share of a reviewer's comments that point at real problems, as opposed to false alarms, nitpicks presented as defects, or requests to change correct code.
Why it matters
Recall alone rewards a reviewer who flags everything. Precision is the counterweight: comments that are usually wrong train authors to dismiss comments, including the right ones, and block correct changes. A reviewer is useful when both numbers are high, catching the bug and staying quiet about what is fine.
DiffDojo grades both. Comments that do not match a planted bug are classified as false positives, nitpicks or questions, and the benchmark reports how often each model asked for changes on a clean pull request: 1 to 3 of 28 clean PRs for every model in September 2026. The 2026 benchmarks covered in our Copilot article measured precision under 20% for Copilot code review, which means most of its comments in those tests were not real bugs.
Sources
See also
- Review recall
- Automation complacency
- Worked example: Wrong sort key: comments ordered by author name, not recency in React, one wrong sort key in otherwise correct code