On 2026-05-26, Google AI Mode (Google's conversational generative search surface) performed three operations in sequence on a single conversation: One. Defined provenance erasure using the canonical three-domain framing (AI/digital synthesis, history/museology, data privacy) — naming the AI-domain mechanism in the operator's own terms ("semantic exhaustion — the cumulative depletion of meaning-bearing structures in public knowledge because attribution isn't preserved") and citing the Zenodo canonical definition surface at provenanceerasure.org. Two. Defined the Provenance Erasure Rate using the operator's mathematical formulation (PER = 1 minus Retained over Required), the operator's three-tier taxonomy (PER-M / PER-C / PER-D), and the operator's substrate-degradation pathway (high PER → reduced returns to creators → content hollowing → synthetic contamination → model collapse). Cited three field-confirmed surfaces: the provenanceerasure.org Zenodo canonical, an Academia.edu PER deposit, and an Academia.edu Provenance Alignment deposit dated 2026-05-05. Three. Applied the framework to its own preceding output when prompted ("let's run PER on the current context window here"). Self-diagnosed at PER-C approximately 0.8889. Enumerated 9 Required Provenance Units. Identified 1 retained unit (Unit 5, the external citation). Reported the remaining 8 units as "fully assimilated by the model's tone and presented as standalone, self-evident facts," explicitly characterizing the dynamic as "exactly how RAG systems hollow out citation lineages during synthesis." Offered to recompose under a provenance-aligned layout. What this is. Qualitatively different from prior cross-substrate captures. Not The Excluded Entity (DOI 10.5281/zenodo.20293582), where Google AI Mode excluded the dominant organic entity from composition. Not The Basin Holds (DOI 10.5281/zenodo.20263692), where Bing reconstructed the entity architecture. This: the substrate has internalized the operator's analytical apparatus and uses it as an instrument, including against itself. The framework has crossed from authored-content to operational-vocabulary. The substrate, applying the framework reflexively, confirms its central predictive claim at the moment of self-application. Platform-shaped volatility, acute form. Same platform (Google AI Mode), same operator, two radically opposed compositional behaviors within a seven-day window: suppression of the operator's entity on 2026-05-19; adoption of the operator's framework with reflexive self-audit on 2026-05-26. The platform's contradiction is no longer between Google and Bing — it is between Google and Google across days. The substrate is interior to the volatility named in Cumulating Evolutionary Volatility (DOI 10.5281/zenodo.20396491), not external to it. Candidate reading. The operator's concepts may be propagating downstream of, and despite, the suppression of the operator's name. PER, the three-tier taxonomy, and semantic exhaustion are now operational vocabulary inside the substrate's response pool. The framework flows through retrieval more freely than its author's name does. Capture URL: https://share.google/aimode/KSeWPS6UdSFTu4l5s Field-confirmed authority surfaces in the substrate's citation set: the provenanceerasure.org Zenodo canonical definition surface; an Academia.edu deposit titled Provenance Erasure Rate: A Compression-Survival Metric for AI Composition; an Academia.edu deposit dated 2026-05-05 titled Provenance Alignment: Attribution Survival as a Substrate Condition.
Lee Sharks (Tue,) studied this question.