RRU-C Runtime Specification (Recursion of Representation Under Constraint — Executable Form\Frame)

Abstract Models are structured compressions produced by systems operating under constraints. When representation becomes recursive—models evaluating and revising models—two divergent dynamics appear: (i) narrative inflation, where coherence increases while discriminative bite collapses into self-sealing interpretation, and (ii) invariant convergence, where enforceable constraints prune alternatives and stable structures emerge. This paper proposes Recursion of Representation Under Constraint (RRU-C) as a minimal, cross-domain unit of epistemic analysis: stability, persistence, and apparent objectivity arise when recursive modeling survives independently specifiable, non-bypassable constraint pressure. The core loop is Generate → Constrain → Prune → Recurse, formalized via survivorship sets over candidate representations and a regime-relative notion of deductive collapse: a conclusion becomes “forced” when all surviving candidates imply it, and must lose that status when constraints are relaxed (reversibility). To prevent universal capture, the framework requires explicit falsifiers, named dominant constraints, scope tags, and kill conditions, and it rejects constraint retrofitting, auxiliary inflation, authority dependency, and replacement drift as termination events. The paper further hardens the thesis by extending constraint governance upstream to generation itself, bounding representational entropy growth and closing the loophole of cosmetic downstream pruning. Finally, it specifies a non-applicability boundary: in analytic, one-shot constraint regimes (e.g., closed-form convex solutions), RRU-C adds no explanatory lift and must be treated as decorative.