A verification act in which a human user independently checks an AI-generated output against an external reference or domain knowledge before relying on it, distinguished from passive acceptance by the presence of an observable confirmation or correction step.
EVENT. One occurrence would be coded when, following an AI output, the user performs an observable checking action against an independent source (cross-reference lookup, recomputation, citation follow-up, or explicit correction of the output) before acting on it. Mere re-reading of the AI text without an external check does not count.
Proposed measurement protocol (not yet empirically validated): Behavioural event coding from interaction logs: count of verification acts per task, with each act tagged by check type (source lookup / recomputation / correction). Two raters double-code a 20% sample; inter-rater agreement could be reported as Cohen's kappa on the present/absent decision.
https://andreasehstandlicenseofclarityloc.github.io/augmanitai-periodic/#topic-meta-reflection