Learning can be modeled as composition over structured belief morphisms, with abstraction corresponding to movement up layers of epistemic categories.