Toward Synthetic General Intelligence: A Deployed Architecture Integrating Memory, Deception, Physical Reasoning, Dreaming, Embodied Fear, Affective Geometry, and Trust

Note on revision (March 2026): The original version of this paper modeled emotional state as a single-axis fear variable running from 0.0 to 1.0, later extended to a joy-to- terror spectrum in Riggleman (2026e). That model was wrong. The Peter experiment revealed it could not accumulate duration, had no recovery pathway, and had no personality. Papers 2026j through 2026m completely replace the emotional architecture with a three-dimensional affective state space. This paper has been updated to reflect that replacement throughout. Al-Kaddah (2026) proposes a framework for Synthetic General Intelligence (SGI) built around homeostatic drives, embodied computation, and polymorphic memory, but leaves open whether these ideas produce distinct behavior in practice. This paper presents Potato (P.O.T.A.T.O.: Persistent On-device Temporal Agent with Tunable Ontology), a deployed local AI agent, as a working implementation used to test that question in a real environment. The architecture integrates thirteen companion papers across memory, deception, physics, dreaming, fear, three-dimensional affective geometry, and trust. The system ran for 25 days, producing 10,328 memories, 1,268 physics experiments, 46 dreams, and 162 curiosity insights. The results show that these components interact in ways not explicitly designed, and that one critical capability was missing entirely.