Input-Attribution Realization
A realization by the user that an unsatisfactory AI output stemmed from an unclear or underspecified prompt rather than from a model error, prompting the user to revise the input. Distinguished from generic insight by its attribution of shortfall to one's own input.
Operational Definition
EVENT. One occurrence would be coded when, following an unsatisfactory AI output, the user produces a turn that (a) attributes the shortfall to their own prior prompt (e.g. 'I wasn't clear', 'I meant') and (b) supplies a more specified reformulation in the same or next turn. The attribution-plus-reformulation turn is the unit counted.
Measurement Schema
Proposed measurement protocol (not yet empirically validated): Rater coding of post-shortfall user turns for self-attribution of the input shortfall and for presence of a more-specified reformulation; agreement would be assessed via Cohen's kappa. Primary metric: self-attribution-and-repair count per 100 user-initiated repairs, optionally cross-checked against measured increase in prompt specificity (added constraints/tokens).