Authority-Level Priors: An Under-Specified Constraint in Hierarchical Predictive Processing

Abstract Hierarchical predictive processing explains adaptive behaviour through precision-weighted inference within a single generative model. Yet explicit belief revision and increased confidence in adaptive self-models often fail to produce corresponding changes in stress reactivity or autonomic regulation. This asymmetry suggests that the framework leaves formally under-specified a governance-level constraint concerning which identity-level hypotheses are permitted to regulate autonomic and behavioural control under uncertainty. We introduce Authority-Level Priors (ALPs) as meta-structural constraints defining a regulatory-admissible subset ${\mathcal{H}}_{\mathrm{auth}}\subset \mathcal{H}$ of identity-level hypotheses. ALPs are not additional representational states, nor hyperpriors over precision; rather, they constrain which hypotheses are admissible for regulatory control, independent of precision-weighted inference operating within that admissible set. Precision determines influence conditional on admissibility, whereas ALPs determine admissibility itself. This distinction offers an explanation for why explicit belief updating and repeated practice readily modify representational beliefs while autonomic threat responses and identity-level predictions often remain stable despite contradictory evidence and sustained effort. We provide a minimal computational formalisation in which policy optimisation is restricted to policies generated by authorised hypotheses, yielding testable predictions concerning stress-reactivity dynamics, recovery time constants, compensatory control engagement, and cross-context behavioural persistence. Neurobiologically, ALPs are hypothesised to manifest functionally through distributed prefrontal arbitration and control networks involved in rule maintenance, conflict monitoring, and top-down modulation. The proposal is compatible with variational active inference and introduces no additional inferential operators, instead formalising a boundary condition required for determinate identity–regulation mapping when multiple high-precision hypotheses coexist. Such a restriction is expressible within existing active-inference schemes; what ALPs specify is a constraint indexed to identity-level hypotheses that is not obligatorily revised by the evidence updating their posterior probability or precision. The central discriminating prediction is that regulatory change following a governance shift should not be anticipated by model-based estimates of posterior probability or precision obtained independently of regulatory measures. ALPs are advanced as an architectural hypothesis to be evaluated through computational modelling and longitudinal stress-induction paradigms.