REASON Framework Uses Two-Loop Mechanism for Robust Learning

REASON introduces a hierarchical two-loop mechanism to resolve residual errors under distribution shifts. A fast 'System 1' loop handles immediate evidence, while a medium-speed 'System 2' loop actively acquires additional information to reduce epistemic residual.