From episodes to populations: evolutionary explanation requires a constructive epistemology

The sciences divide into those that discover laws and those that reconstruct histories. We argue that this division does not reflect a difference in subject matter, but a difference in epistemic regime. Law-based sciences operate under episodic closure: systems are idealized so that the outcomes of prior interactions do not alter the rules governing future ones. This regime-defining idealization (distinguished from pragmatic idealization) underlies the predictive successes of physics, but creates a systematic blind spot for evolutionary dynamics. We formalize this distinction using Stability-Driven Assembly (SDA), a minimal non-equilibrium framework in which differential persistence couples episodes into population-level evolutionary dynamics without genes, replication, or predefined fitness functions. Representing compositional objects as λ-calculus terms, we show that episodic science studies isolated λ-reductions under fixed rules, while evolutionary science studies populations of λ-instantiations whose outputs re-enter the space of operators. The resulting dynamics are self-modifying and irreducibly sequential: each step rewrites the conditions for the next. A four-quadrant taxonomy locates episodic science, evolutionary science, and two commonly conflated intermediate cases: formal possibility and constructive potential, within a single framework. From this analysis we derive the "No Free Telos" constraint: in constructive systems where population feedback reshapes the effective dynamics at each step, the cost of predicting future states cannot in general be reduced below the cost of simulating the generative history. The resulting framework bridges episodic and historical sciences, not by reducing one to the other, but by identifying population-level memory as the structural condition that transforms law-governed episodes into open-ended evolutionary processes.