A coherent engine that needs less of the model the more it learns.

It confirms only what closes from math against constants of truth — and publishes its own edges. Every question it can't answer becomes a fixed rule, so its dependence on the statistical model falls over time. The opposite of the scaling paradigm.