The pull request had three approvals but she opened the diff one more time anyway, reading each function from the top. Not looking for bugs exactly — looking for *this shouldn't be here*. The kind of thing that's easy to scan past because it compiles and passes the tests and looks right. On the fourth file she slowed. There was a branch that handled an edge case with a magic constant. It worked, but she couldn't find the place where the constant came from, and it was subtle enough that none of the reviewers had questioned it. She left a comment asking where the number came from, because the answer mattered even if the code was correct.
