Regulatory Intelligence Program

The Regulatory Intelligence (RI) Program is a research program advancing a viability-first paradigm for artificial cognition, in which intelligence is defined as a system’s capacity to maintain internal coherence, bounded dynamics, and recoverability under stress, rather than to optimize external task performance. The program reframes cognitive architectures as regulated dynamical systems with explicit internal physiology. Stability, not accuracy, is treated as the primary design objective. Task competence is understood as an emergent strategy for preserving viability, rather than as the goal of cognition itself. This record synthesizes the full Regulatory Intelligence program, integrating foundational theory, architectural principles, empirical findings, and falsification protocols. The program is implemented and empirically validated through SpiralBrain v3.0, a non-learning neurosymbolic cognitive system designed as a scientific instrument rather than a benchmark-optimized model. SpiralBrain operates on a bounded geometric manifold, exposes internal physiological variables (coherence, drift, hazard, affective state, phase relationships), and enforces stability through an explicit regulatory core. Key contributions of the RI Program include: A formal definition of intelligence as viability under cognitive and environmental stress. A strict non-learning constraint, enabling clean falsification, reproducibility, and separation of regulation from plastic learning. Geometric homeostasis over a bounded cognitive manifold. Elastic adaptation within runs without cross-run parameter persistence. A reproducible phase-lock stability region governing multi-pathway cognitive coordination. A bifurcated evaluation framework separating cognitive physiology from task accuracy. Empirical stress-testing protocols that treat benchmarks as stressors, not optimization targets. The RI Program positions regulation as a first-class architectural commitment and proposes Regulatory Intelligence as a stability substrate beneath adaptive or learning-based systems. Rather than replacing learning-centric AI, RI provides a foundation for building stable, interpretable, and intrinsically aligned cognitive systems suitable for scientific study and safety-critical environments.