Universe Routing: Why Self-Evolving Agents Need Epistemic Control

TL;DR

This paper introduces the universe routing problem, emphasizing the importance of epistemic control in self-evolving agents, and demonstrates that specialized, modular routing improves reasoning accuracy, efficiency, and lifelong learning.

Contribution

It formalizes the universe routing problem, shows that hard routing matches soft MoE accuracy with greater speed, and demonstrates the effectiveness of modular architectures for lifelong learning.

Findings

Hard routing matches soft MoE accuracy, 7x faster.

A 465M-parameter router reduces generalization gap by 2.3x.

Rehearsal-based continual learning achieves zero forgetting, outperforming EWC.

Abstract

A critical failure mode of current lifelong agents is not lack of knowledge, but the inability to decide how to reason. When an agent encounters "Is this coin fair?" it must recognize whether to invoke frequentist hypothesis testing or Bayesian posterior inference - frameworks that are epistemologically incompatible. Mixing them produces not minor errors, but structural failures that propagate across decision chains. We formalize this as the universe routing problem: classifying questions into mutually exclusive belief spaces before invoking specialized solvers. Our key findings challenge conventional assumptions: (1) hard routing to heterogeneous solvers matches soft MoE accuracy while being 7x faster because epistemically incompatible frameworks cannot be meaningfully averaged; (2) a 465M-parameter router achieves a 2.3x smaller generalization gap than keyword-matching baselines, indicating semantic rather than surface-level reasoning; (3) when expanding to new belief spaces, rehearsal-based continual learning achieves zero forgetting, outperforming EWC by 75 percentage points, suggesting that modular epistemic architectures are fundamentally more amenable to lifelong learning than regularization-based approaches. These results point toward a broader architectural principle: reliable self-evolving agents may require an explicit epistemic control layer that governs reasoning framework selection.