Capability-Limit Encounter
The moment a user encounters the limits of an AI system: the system cannot solve the task, returns incorrect information, or exposes a comprehension boundary, giving the user direct evidence of what the system cannot do.
Operational Definition
EVENT. One occurrence is coded when, within an exchange, the AI fails on the user's request in an observable way (explicit inability, a detectable error, or a stated comprehension limit) and the user registers this limit in a subsequent turn. The user's recognition turn following the AI limit is the countable marker.
Measurement Schema
Proposed measurement protocol (not yet empirically validated): Failure-event rater coding: coders annotate AI-limit events (refusal/inability, factual error, comprehension limit) and the user turn that registers them; count limit-recognition events per session and categorize by limit type; inter-rater agreement would be assessed via Cohen's kappa on event type.