Whether an AI model's stated reasoning actually explains how it arrived at its answer, or if it's post-hoc justification.