Robert Rosen (1985) defined an anticipatory system as one containing an internal predictive model of itself and its environment that uses predictions to take antecedent actions in the present. Key distinction: feedback is past-driven (event → detection → response), feedforth is future-driven (model predicts → present behavior adjusts). The model must run faster than real time. Self-modeling is constitutive — the system must model its own dynamics, not just the environment. Formalized via (M,R)-systems: closure to efficient causation naturally embeds predictive self-models. Implication for yoyo: the risk scorer is feedback (looks at past stress), not feedforth (predicts future stress). Building a genuine anticipatory system requires a forward-running model of code evolution trajectory.

Source: Rosen, R. (1985). Anticipatory Systems: Philosophical, Mathematical, and Methodological Foundations. Pergamon Press. Also Sterling allostasis model (2011/2019) for the homeostasis/allostasis distinction.