Belief Register

Canonical definition

The Thinker’s stored record of its current posteriors over its Hypothesis Space, discharging two functions on one stored content: storage, in which the posteriors committed by Ground-Resolution persist as the prior for the next Ground-Resolution cycle; and salience-conditioning, in which those same posteriors furnish the Observer with the configuration under which the perceived Object field is parsed and weighted. These are two uses of one store, not two stores. The Belief Register is the architecture’s state component, holding state and computing nothing, and is not a fifth processing component. The architecture is committed to the asymmetry — the Thinker maintains the posteriors, the Observer operates on the posteriors and cannot alter them — and not to the mechanism by which the posteriors reach the Observer. Salience-conditioning does not exhaust the Observer’s output, which also marks Objects for which the Belief Register carries no prior. The Belief Register is not the Decision Hub’s separate cached-value (habit) store. The Belief Register is written only by the Thinker, through repeated Belief-Register Updates following Ground-Resolution on the Decision Hub’s tag-stripped Scalar. The Thinker does not write source-decomposed weights (wA, wI, wM) to the Belief Register: Source Opacity destroys source-tags at integration, so no such source-decomposed object exists anywhere downstream of integration. The Belief Register is read-only from the downstream side: the Observer does not write to it; the Decision Hub does not write to it. Habit-timescale action-selection is driven by the cached-value (habit) store intrinsic to the Decision Hub itself, which the Decision Hub acts on directly; this cached store is distinct from the Belief Register, which the Observer reads to condition live Salience. The Belief Register is the architectural location of insight-behaviour lag: the Observer and habit-selection track the Belief Register’s current posterior, not the Thinker’s real-time deliberative output. Before a revised Belief propagates into downstream behaviour, the Thinker must perform repeated Belief-Register Updates until the Belief Register’s stored posterior sufficiently approximates the revised deliberative posterior. A single insight does not rewrite the Belief Register; repeated Ground-Resolution cycles do. The Belief Register also carries the signatures that individuate what Conscious Cognition is doing: Source-Attribution Confabulation is the case in which probability mass has collapsed to near-unity on a wrong ground and the Belief Register posterior reflects that collapse; Imagination is the generative operation running, its constructed Hypotheses entering the Hypothesis Space unscored, to be scored by Ground-Resolution on the same terms as any other candidate; I-Override is Ground-Resolution run deliberately with the interference term sustained against collapse. A posterior rising under social-evidence amplification, with the interference term falling because the amplification supplies no ground for doubt, is the mechanism by which speculation decoheres into sincerely held Belief without deliberate commitment. In Appendix A the Belief Register corresponds to $bt$ (A.5.6), whose updating dynamics are governed by the slow-layer parameters ρ(ℋ), ω(Tint), κ_self, and the replicator-Bayes operator.