The nativist–empiricist debate has long organized cognitive science, framing cognitive mechanisms as either innate or learned. Recent advances in controlled rearing and scalable domain-general neural networks now allow this debate to be tested with unprecedented rigor, by raising newborn animals and deep neural networks under tightly matched developmental conditions. Strikingly, when pushed to this empirical limit, the debate does not resolve: the same developmental phenomena support both radical empiricism and radical nativism. We argue that this stalemate reflects a conceptual error. The debate conflates explanatory levels and ontological orientations, with nativists emphasizing stable symbolic structures and empiricists emphasizing the subsymbolic processes that produce those structures. We propose dissolving the dichotomy within a multilevel framework of cognitive development.