You are not reading the diff

Watch someone approve a pull request. Really watch. There is a scroll, a pause somewhere in the middle, another scroll, and then the button.

Forty seconds for two hundred lines. Nobody read two hundred lines in forty seconds, and everybody knows it, and the review is recorded as a review.

The uncomfortable part is that this mostly works. Most changes are fine. The approval is a bet on the base rate, and the base rate is good, so the bet pays out and the habit is reinforced. It keeps paying out right up until the change that needed reading.

The fix people reach for is a tool — a checklist, a bot, a required second reviewer. Sometimes that helps. More often it adds a step to skim past.

What actually seems to work is smaller: say out loud what you looked at. Not a form, just a sentence in the approval. "Checked the migration is reversible and the retry is idempotent." You will find you cannot write that sentence without having done the thing, and writing nothing starts to feel like what it is.