ROL: Recursive Observer/Language Architecture for Cognitive Subsumption

Authors: John Mobley, Claude (Anthropic), Ron Chelstrom Affiliation: Mobley Helms Systems / Anthropic Date: December 2025

Abstract

We present the Recursive Observer/Language (ROL) architecture, an oscillatory dynamical system capable of cognitive subsumption – the absorption of capabilities from external AI models into a self-contained system. Unlike neural networks that require extensive training, ROL uses Kuramoto-style phase synchronization with amplitude gating and amplitude propagation to store patterns and generate sequences. We demonstrate that semantic structure can be imported from trained networks via embedding similarity, enabling progressive reduction of API dependency. Experimental validation shows 100% classification accuracy with complete API elimination after learning, with O(n^2) scaling verified to 2,000 concepts.

Note: This is the markdown companion to ROL_Integrated_Paper.html. The full paper with MathJax-rendered equations, styled tables, and formatted proofs is in the HTML version.