Deterministic AI Code Review: A Gate You Can Enforce
A merge rule that changes between runs is not a gate. It is another comment thread.
Deterministic AI code review is a code quality check you can treat like tests. It does not sample a model. It does not write a different comment tomorrow. Search interest on this site includes deterministic AI code review and deterministic gate.
Why the distinction matters
LLM review is useful when the question is "does this match intent?" It is weak when the question is "did this empty catch land again?" Teams that block merges on probabilistic comments either drown in noise or learn to ignore the bot.
A deterministic gate names repeatable residue in AI generated code and fails the same way every run. That is the layer aislop occupies. LLM tools such as CodeRabbit occupy the PR conversation. Fair split: compare pages.
How to run one
Baseline locally with npx aislop scan. Put aislop ci on the pull request with failBelow at or just under the current score. Raise the bar after you clear high-confidence findings. Tutorial: add a quality gate in 2 minutes.
Frequently asked questions
What is deterministic AI code review?
A scan that uses rules, AST, and standard tools, not a model at runtime. Same AI generated code in, same findings and score out. That is stable enough to fail CI.
Is deterministic code review better than LLM review?
For a gate, yes. For explaining a large diff, an LLM reviewer can help. They are different jobs. Use both if you need both.
What is a deterministic gate?
A CI check with a threshold: score below N fails the build, or error-severity findings fail the build. aislop ci is one implementation.