Code review examines a proposed change before it merges. It is a gate for correctness and maintainability, but also a place to see how developers actually work. A review can catch a defect today and reveal a standard that needs improvement tomorrow. section 4.4.3
Ask questions that can change the merge decision:
- Does the change implement the spec without adding unrequested behavior, and do its tests pass? section 4.4.3
- Does error handling, logging, file structure, and testing follow documented team patterns? section 4.4.3
- Are new dependencies justified, coupling limited, and the code isolated enough to remove later? section 4.4.3
Block a change for failed tests, spec violations, or consequential departures from documented standards—not a reviewer's personal brace or naming preference. Ask whether a comment prevents a defect, improves readability for the next maintainer, or clarifies intent. Repeated justified departures across pull requests are a reason to investigate and update the standard, not enforce it blindly. section 4.4.3 section 13.2
A review that discourages improvement
A developer replaces a hand-rolled cache with a tested library, removes 60 lines, and adds property-based tests. A reviewer spends four rounds on personal style preferences and untried alternatives. The next change may be safer to propose but less valuable: a small fix instead of another improvement. Review the library's justification and the change's behavior; keep preferences nonblocking. section 13.2
AI-generated code needs the same gate even when CI passes. The person who reviews and merges it takes responsibility for understanding what it does, why it works that way, and whether a future maintainer can change it. Named constants, explicit business rules, and tests help preserve that understanding. section 13.1.1











