Selection mechanisms in open adaptive systems remain fragmented across optimal control, active inference, basal cognition, and dynamical systems theory. Each tradition asks how a system chooses what to do next, and each answers in vocabulary that does not translate cleanly into the others. We argue that a useful path forward results from abandoning the goal-first framing in favor of a substrate-neutral, restraint-first formulation in which continuation is selected by feasibility under restraint rather than by objective optimization. The framework, named RCOT (Restraint, Continuity, Observation, Translation), proposes an executable selection rule. Within RCOT we define "locked continuation" as a state in which coupling depth D has been driven past a criticality threshold τ relative to a prior basin and that basin's volume in the feasible set F has collapsed to zero. Locked continuation provides a candidate mechanism for fold-like irreversible transitions, a quantitative early-warning signature for regime change, and a non-teleological account of agency as feasibility maintenance under accumulating restraint. The construct connects to Friston's free energy principle, Levin's agential material program, Aubin's viability theory, Thom's catastrophe theory, Bickhard's interactivist process metaphysics, and Kauffman's adjacent possible. Implications include constraint-native AI architectures that resist goal misgeneralization, candidate predictions for which Levin-lab phenomena should display D/τ signatures, a common geometric framing for fold catastrophes across scales, and operational early-warning indicators for ecological, market, and cellular regime change. The paper develops the operator, situates it across six adjacent literatures, surveys seven candidate empirical domains for which it predicts measurable D/τ signatures, and identifies the open empirical and theoretical questions that organize a research program around constraint-aware selection.
Ricardo Cruz Orozco (2026) studied this question.