Most agent failures are data failures dressed up as model failures. The audit catches them before deployment. Twenty questions across five dimensions with concrete evidence requirements for each YES answer.
Companion: filled sample. A completed audit scoring 31/40 (78%) with a single remediation item and named owner.