Anatomy of an evaluation: five parts, one decision

Build lab: write the five-part eval spec for a refund assistant — task, dataset, grader, metric, decision rule — then watch the fifth part be the only one that stops a build. The team's spec reports 70.0% in Slack every Monday while all 7 builds ship; adding one threshold on one slice stops 5 of them, and an aggregate-only floor still lets 2 through.

This Evals Engineering lesson is part of a paid plan. Sign in with a plan that includes this course to read it.