A Dual-Model Architecture for Cognition

This paper introduces Ze, a novel theoretical framework for cognitive architecture based on the concurrent operation of two distinct generative models of the environment: a causal (forward) model M_A and a counterfactual (inverse) model M_B. The core dynamics of Ze arise from the minimization of two separate variational free energies, F_A and F_B, and the management of the conflict between them, ΔF = |F_A - F_B|. This conflict regulates a phase transition between an interference regime, where model outputs are constructively fused, and a localization regime, which resolves the conflict through a discrete projection. We formally establish a deep structural isomorphism with quantum measurement, particularly the double-slit experiment, without invoking quantum physics in the substrate. Ze is proposed as a complete, falsifiable theory that reinterprets cognitive "collapse" as an optimization-driven transition, generates novel experimental predictions, and integrates perception, action, and representational learning into a unified architecture. This preprint provides the full mathematical elaboration of the framework.