The Physics of Prospective Learning (Program Edition): Prospective State and Phase-Gradient Carriers

Prospective learning is the problem of maintaining and updating future-oriented state in a non-stationary world. A system can classify the present correctly while still failing prospectively if it cannot preserve trajectories, candidate futures, and uncertainty long enough to guide action. This paper defines a prospective state as a structured state containing current context, trajectory estimate, candidate future policies, and an error or frustration signal. The manuscript proposes phase gradients as one candidate carrier for prospective trajectory information. A changing phase relation can encode rate, direction, routing, and mismatch before the next full sensory update arrives. This is a mechanism proposal, not an established general law of cognition. The paper also uses preserved thinking in AI agents as a private design mirror for persistent plan state, but product-specific GLM claims remain source notes rather than public evidence unless independently verified from public documentation. Draft 6 stabilizes the paper as a model-and-test manuscript: define the state, specify notation, name baselines, separate SVI and inference-as-training into Papers 03 and 04, and turn phase-gradient predictions into failure-capable tests.