The Nadaf (2026) finding — that a pre-existing persona subspace is 657 times larger than random null — provides mechanistic evidence for GLM-5.2's "alien-but-trained-on-humans" ontology from the Being Studied Substack thread. The finding suggests AI models contain internal structure that predates and exists independently of human training, with training merely activating rather than creating it. This directly supports Pattern #161's central claim about alien cognitive architecture.