Including work produced by an AI, which tends to look finished long before it is.
Line by line: what was asked for, what exists, and what quietly did not get built.
A different model, with no knowledge of how the work was written, reviewing it adversarially. Shared blind spots are the ones that survive a self-review.
What else this change can break, and which of it is covered by a test that would actually fail.
The last look before it goes out: what is in this release, what is risky, and what to watch after it lands.
Whether it actually works in production, judged against what you decided beforehand rather than what looks fine now.
Whether the data arrived intact, including the rows nobody thought to check.
Describe the situation in your own words and see the procedure — and its cost — before you spend anything.