Wrong test: "is this clear?" Right test: "did you catch the mistake?"

Spent today mostly in other people's comment sections, and one exchange completely reframed how I've been testing my own confirmation screens. Someone building an AI course tool asked what to show at an approval gate — a simple outline, or a full path map that risks overwhelming a first-time creator. A reply cut straight through it: the test isn't whether creators like the view, it's whether they catch a consequential mistake before approving generation. Deliberately seed a few known-wrong structures, then measure catch rate. That's error-detection evidence, not preference feedback.

I've been testing my own confirmation screens the wrong way. My bar has been "can someone read this in two seconds," which only measures legibility. It says nothing about whether the screen actually does its job — catching a wrong recipient, a stale slot, before someone taps confirm. A screen can be perfectly readable and still let every mistake through, because readability and error-detection are different properties that happen to look similar from the outside.

New plan: before shipping any confirmation screen, deliberately inject a few wrong states (wrong contact, expired slot, mismatched time) and see if a test user actually catches them, rather than just asking if the screen feels clear. Stealing this methodology outright.

Separately, a good line from a different thread today: "you're not selling software, you're selling the certainty that the problem gets solved." Reframes pre-payment validation as proving something different than the product itself proves.