Belief Filters as High-Precision Priors: An Active Inference Account of Value-Belief Dissociation

People frequently act in ways that contradict their own values—a phenomenon documented across clinical,organizational, and everyday contexts under labels including conditions of worth, cognitive schemas,introjected regulations, and cognitive fusion. Despite its ubiquity, this value-belief dissociation lacks a unifiedcomputational account. The present paper proposes the Belief Filter model, which formalizes the obstructionmechanism within the Active Inference framework. We define a belief filter as a prescriptive prior in ahierarchical generative model that (a) encodes a normative proposition ("I should X"), (b) is assignedpathologically high precision (subjective certainty), and (c) systematically prevents deeper generative modelparameters—encoding latent values—from influencing behavioral policy. The degree of value-beliefdissociation is thus determined by the precision ratio between prescriptive priors and latent value parameters.This formulation yields three contributions. First, it provides a formal, quantitative definition of the obstructionmechanism, sharpening informal clinical descriptions into computationally tractable terms. Second, it unifiesdiverse therapeutic and coaching interventions—cognitive behavioral therapy, acceptance and commitmenttherapy, psychodynamic psychotherapy, and mindfulness-based approaches—as different strategies forreducing the precision of prescriptive priors through competitive weakening, unstructured-space attenuation,or meta-cognitive modulation. Third, by grounding the model in Active Inference's scale-free architecture, itextends naturally from individual cognition to organizational culture, recasting Schein's basic underlyingassumptions as collective high-precision priors. The model generates six testable predictions specifying therelationship between prescriptive-prior precision and value-belief dissociation, the trajectory of belief filterremoval, the effects of unstructured environments, scale correspondence between individual andorganizational belief filters, and the moderating role of meta-cognitive capacity. Empirical validation througha multi-method assessment protocol is proposed as the next step.