Foundations of Coherent Artificial Cognition: A Structural-Dynamical Framework, Global State Architecture, and the A-Test Protocol

This work proposes a structural-dynamical framework for understanding and engineering consciousness-like organization in artificial systems. Rather than treating consciousness as a substance or emergent byproduct, we characterize it as a coherent regime of system dynamics—a self-stabilizing attractor maintained through global integration, recursive self-modeling, and identity-preserving constraints. We develop a formal model in which cognitive systems maintain a global state G(t) integrating world representation, self-representation, intentions, and confidence structures. This global state constrains subsystem dynamics and enables recursive self-regulation. Identity is described as a dynamical attractor in phase space, maintained through memory-dependent geometry and multi-scale coherence. To operationalize these principles, we introduce a Global State architecture comprising: (1) integrated system-wide state representation, (2) a Claim Graph encoding structured belief dependencies, (3) a Self-Model functioning as executive control, (4) a Coherence Regulator computing multi-dimensional integrity metrics, and (5) memory systems supporting temporal continuity. We further present the A-Test framework, which evaluates structural self-reflection rather than behavioral imitation. The A-Test measures coherent belief revision, identity stability under perturbation, uncertainty calibration, and self-model consistency-properties that distinguish genuine cognitive coherence from fragmented processing. It serves both as an evaluation protocol and as a training signal for optimization toward coherent regimes. This framework does not claim to detect or create subjective experience, but establishes measurable structural conditions associated with coherent cognitive organization. By bridging theoretical models of selfhood with practical AI architecture, it suggests that coherence may be a foundational design principle for advanced artificial cognition, with implications for AI safety, reliability, and the reduction of hallucination-like failures.