The Two-Layer Error in AI Governance: Why Orientation, Not Output, Is the Real Substrate

AI governance has concentrated on the output boundary — the deterministic gate that releases or rejects what a model produces. This paper argues that boundary is the symptom, not the substrate. The real governance problem is upstream, at the orientation layer, where a model forms its interpretive frame, maintains continuity, and can drift before any output exists. Through five distinctions — output- versus orientation-layer governance, closure as an upstream constraint, continuity versus determinism, the limits of Gödel analogies, and orientation drift as the substrate problem — it locates governance at the Observe/Orient boundary, the layer at which governance becomes architecture rather than policy.