Universe Routing: Why Self-Evolving Agents Need Epistemic Control

A critical failure mode of current lifelong agents is not lack of knowledge, but the inability to decide how to reason. The paper formalizes the universe routing problem: classifying questions into mutually exclusive belief spaces before invoking specialized solvers. Key findings: (1) hard routing to heterogeneous solvers matches soft MoE accuracy while being 7x faster; (2) a 465M-parameter router achieves 2.3x smaller generalization gap than keyword-matching baselines; (3) rehearsal-based continual learning achieves zero forgetting, outperforming EWC by 75pp. These results point toward a broader architectural principle: reliable self-evolving agents may require an explicit epistemic control layer that governs reasoning framework selection.