Affective Endurance and Criticality in Adaptive Agents: A Dynamical Systems Theory of Emotional Governance

This paper formalizes emotion as a structural governance necessity for adaptive intelligent systems operating in non-stationary environments. We introduce the Affective Governance Model (AGM), a dynamical systems framework built on four core contributions: (1) an Affective Endurance variable H(t) representing a finite governance resource whose exhaustion produces bifurcated collapse into Freeze (rigidity) or Runaway (instability) regimes; (2) the Emotional Criticality Condition (ECC), a formal inequality Λ(t) ≥ Θ that predicts affective phase transitions from measurable internal state variables; (3) a middle-layer interpretive buffer that mediates between perturbation and action as a governance saturation signal; and (4) a demonstration that the complete system exhibits self-organized near-criticality through a discrete event drive–release architecture, yielding testable predictions including power-law episode statistics, critical slowing down, and finite-size scaling. Toy simulations confirm all six falsifiable predictions. The framework bridges affective computing, reinforcement learning, and statistical physics, offering a principled alternative to static gain-scheduling approaches in adaptive control.