Humans are good at sifting valid feedback from bad feedback. But we are bad at spotting subtle bugs in PRs.