Our review queue had a familiar shape: half the comments on every pull request were things a machine could have said. Unused imports, a hardcoded token in a test, a missing timeout on an HTTP call, a new endpoint without a test. Reviewers spent their attention on those and skimmed the parts that needed a human, like the data model change. We built a local review gate that runs before git push: fast deterministic checks first, then security scanners, then the test suite in Docker, and finally an...