Computational psychiatry has reframed psychiatric symptoms as configurations of inference and control variables — prediction errors, precision weights, and learning rates — but has largely treated these parameters as biologically unconstrained. Here we propose a constraint-based framework in which psychiatric phenomena arise from a shared predictive-control architecture operating under explicit biological and computational constraints. The framework's central claim is that viable, conscious-capable brain dynamics require the joint satisfaction of two constraints: sufficient metabolic support and sufficient large-scale integration. Together these define a two-dimensional state space within which psychiatric configurations must lie, while configurations violating either constraint fall outside the regime in which the framework applies.Within this constrained space, behavior is organized by viability-weighted approach–avoidance tendencies extending through cross-modal predictive control into hierarchically abstract domains. Four levels are distinguished: (1) metabolic–integrative constraints, indexed by FDG-PET cortical metabolism and perturbational complexity (PCI); (2) viability-weighted tropisms anchored in minimal gradient-sensing systems; (3) cross-modal predictive control involving prediction-error signaling, precision weighting, transition uncertainty, and persistence dynamics; and (4) hierarchical abstraction, including anticipation, social inference, and recursive self-modelling. Psychiatric disorders are interpreted as parameterized configurations of these shared processes rather than discrete diagnostic entities.We apply the framework across eight clinical domains and show that each maps onto characteristic alterations in uncertainty processing and control dynamics. PTSD and OCD are presented as contrasting abstraction-level configurations: PTSD as excessive stabilization of high-precision threat priors and OCD as failure of uncertainty resolution under elevated transition uncertainty. The framework generates interaction predictions linking metabolic and integrative state to disorder-specific precision dynamics, differentiating it from predictive-processing accounts that do not specify explicit biological constraint boundaries.
Shivashanmugam et al. (Tue,) studied this question.