✶ Epistemic Akrasia and the Fear of Updating: Why Perceived Evidence Fails to Move Assigned Confidence

The Credencing framework evaluates whether assigned confidence answers to evidence. Its existing papers distinguish Objective Evidence, E0, Perceived Evidence, EP, Assigned Credence, CA, and Deep Rationality, SD; separate evidence-processing failures from belief-integrity failures; defend warranted uncertainty; diagnose rationalization; scale credencing to institutions and AI-mediated inquiry; distinguish raw from excess gaps; make posteriors answerable to evidential history; credential frames and categories; distinguish compatibility from confirmation through likelihood literacy; discipline priors through base-rate literacy; and separate evidential support from explanatory satisfaction. This paper adds a missing interior layer: epistemic akrasia. Agents sometimes do not merely misread evidence or lack inferential skill. They recognize, with some degree of meta-awareness, that confidence should move, but experience the update as costly, frightening, humiliating, identity-disruptive, relationally dangerous, or practically destabilizing. The lacuna is therefore the lived and formal structure of the EP-CA gap from the inside: the moment in which an agent senses update pressure but cannot yet become the kind of person who believes differently. The paper formalizes meta-recognition, update pressure, update cost, update affordance, warranted update latency, akratic lag, update drag, update compliance, public-private avowal gaps, and staged repair. It argues that epistemic akrasia is not identical to ignorance, ordinary uncertainty, self-deception, strategic lying, or rationalization, though it can interact with all of them. Its practical result is an akrasia audit: identify the evidence, estimate the agent's perceived evidential pressure, distinguish recognition from access failure, map the anticipated costs of updating, grant warranted latency where safety and identity repair require time, diagnose excess akratic lag only after that slack is considered, and design staged update paths that lower the cost of belief integrity. Credencing thereby gains a humane account of why evidence sometimes reaches the agent before the agent can follow it.