The Salience Engine: A Coordination Architecture for Salience-Weighted Mode Dynamics

This paper introduces the Salience Engine, a coordination-level architecture proposing that cognitive coordination can be formalized as a five-stage pipeline: Pressure, Parsing, Salience, Mode, and Action — which together describe how relevance-weighted signals guide transitions between metastable cognitive configurations. The model is conceptually grounded in predictive processing and the Free Energy Principle while remaining compatible with dynamical systems approaches and attention-based computational architectures. Cognitive coordination is modeled as motion through a salience-structured state space in which precision-weighted signals influence transitions between metastable processing modes. A precision calibration layer — jointly regulated by neuromodulatory state, interoceptive input, learned priors, and active goal context — determines the gain of salience computation and provides the mechanism through which individual differences, clinical variation, and intervention effects arise. Neurodivergent conditions such as autism, ADHD, and trauma are modeled as parametric variations in salience weighting and mode stability rather than as deficits in cognitive capacity. Executive function is framed as meta-cognitive access to one's own salience dynamics, enabling deliberate intervention in otherwise automatic processing. The framework generates empirically testable predictions regarding neural coordination dynamics, cognitive switching costs, and individual differences in salience weighting, and is proposed as a coordination-level scaffold that integrates existing cognitive theories while providing a structured vocabulary for describing relevance-driven cognition across biological and artificial systems.