The verifier owns done

Repair lab: a harness decides done from the model's own words and certifies a wrong answer without opening a file. Rewire decide() so the harness reads the artifact the task declared and byte-compares it with an independent oracle; five scripted agents show why agent-named paths, agent-supplied expectations and fuzzy matches let lies through.

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