The Closure Constraint as a Generative Frame: Orientation Drift in Closed Cognitive Systems — A General Result, a Status Taxonomy, and Pre-Registered Predictions

This paper states a general structural result and then does what a post-hoc account cannot: it predicts. The result — the closure constraint — holds that a closed cognitive system cannot maintain its orientation under perturbation by means of a corrective implemented within the system’s own dynamics, because that corrective is governed by the dynamics it must resist and decays under the same perturbations; stabilization therefore requires an external substrate. Several documented architectural failure modes in transformer-based systems are instances of this single structure. The paper consolidates the result, defines a status taxonomy — resolved, open, predicted — for organizing instances against it, states the boundary of the class so the claim remains falsifiable, and pre-registers two theory-derived predictions of failure modes the constraint implies but that have not yet been documented. Each prediction is labeled as a hypothesis, derived explicitly from the result, and given a falsification condition. The predictions concern requirements, not mechanisms: they state what an external substrate must supply, not how any particular substrate does so.