Code Reviewer

Ask for a code review and you get forty findings. Some are real, most are style, and a few are invented. Nobody acts on a list of forty, so the list gets saved and never opened, and the review was theatre.

Code Reviewer is a loop rather than a review, and the second step is the one that matters. Before a finding is allowed onto the report it has to be proved: write the actual input that triggers it and the actual wrong output it produces. A finding nobody can trigger is a guess, and guesses are thrown away rather than softened into maybes.

1. 01Sweep the codebase once per dimension. Correctness, then dead code, then accessibility, then error handling, and so on. One pass looking for everything finds nothing.
2. 02Prove each finding with a concrete failure. Discard the ones you cannot trigger.
3. 03Rank what survives by what it actually costs, not by how easy it was to describe.
4. 04Fix the top one. Make the smallest correct change. Run the check that proves it.
5. 05Do not start the next one until this one is proven. Never batch fixes.
6. 06Sweep again to confirm the fix introduced nothing new.

It stops on its own. When the list is empty it says so, and when something left needs a decision that is yours rather than the machine's, it hands it back instead of guessing at what you would have wanted.