Most AI governance debates begin at the threshold—the point where model behavior seeks to become record, route, claim, or institutional fact. Threshold governance is necessary. It is not the beginning of the architecture. This paper argues that the correct governance chain is substrate → primitives → threshold, in that order and only in that order. The substrate defines the physics of the meso-regime in which LLMs operate. Primitives must be built as binding rules under substrate uncertainty, not as abstractions that assume stability the substrate does not possess. The threshold enforces consequence only when primitives that correctly account for substrate illegibility are in place. Governance collapses when any link in this chain is skipped—not through implementation failure, but through architectural error.
Narnaiezzsshaa Truong (Sun,) studied this question.