The risk log says the support classifier may be unfair, but the team has no test that could prove the risk is controlled or force a launch change. Risk -> harm -> population -> metric -> segment -> threshold -> owner -> action rule. Common trap: use one average accuracy number and call it fairness evidence. Averages can improve while a relevant subgroup gets worse. Name the harm Define the concrete failure: urgent complaints from non-native English speakers may be misrouted to standard support. A test plan starts with a consequence, not a vague fairness word. Choose metric and segments…
Read the full lesson
Sign up free: one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in