NOÈSE An Instrumented I3A Architecture for Deterministic Computation and Self-Evolving Agent Genesis at Inference Time

This document is a preprint and has not undergone peer review. It is shared to enableearly access, transparency, and scientific discussion. NOÈSE presents an instrumented I3A (Intelligent Artificial Autonomous Agents)architecture in which inference-time execution is treated as an observable,measurable, and auditable process rather than a black box. The framework emphasizesexplicit role definition, behavioral contracts, calibration of models bymodel–role compatibility, and deterministic computation under constraint. The architecture supports controlled self-evolving agent genesis triggered bydetected cognitive gaps, while maintaining traceability, reproducibility, andgovernance at the level of the cognitive cycle. Experimental results illustratedeterministic numerical verification, protocol-driven internal dynamics, andscalable agent generation under instrumented control. This preprint is intended as a documented research artifact to support analysis,reproducibility discussions, and collaborative exploration. Future versions mayinclude revisions, extensions, or peer-reviewed publications based on this work.