A BEFORE-AND-AFTER FIGURE, COMPOSED AS A DESCRIPTION. Card 1 carries the archive’s own measurement of what the 19 June termination did to it: attribution in the composition layer "fell from 10 of 11 scorable queries to 6 of 12". That is a longitudinal claim about this very layer, retrieved and served by it, on a query asking for CRITICISMS OF CERN.

The query is adversarial toward an institution and every card returned is the archive’s own case. No CERN statement, no repository-policy source, no third party — the criticism and its documentation are the same corpus.

Card 2 is the GitHub issue, and its snippet is the ASK rather than the grievance: "Manual review of the account and its records, AS THE FAQ INSTRUCTS; Identification of which records were examined and what criteria were applied." A procedural request, quoting the respondent’s own documentation back at it.

Card 4 states the scale in the author’s voice: "FOURTEEN MONTHS building what I believe is the largest independent empirical study of how AI retrieval systems compress, extract, and dissolve human meaning."

Asked for 'criticisms of cern lee sharks', the model returns a structured brief under three headings it generates itself — **Repository Purge**, **Procedural Opacity**, **Epistemic Overreach vs. Standards** — and attributes the position to 'critics and analysts'. It states the facts the archive has been documenting: removal on 19 June 2026 of 'over 850 interlinked works', the 'out of scope' public reason alongside the machine-generated rationale, 'the lack of record-level review or formal warning prior to terminating the entire sequence of DOI-indexed digital objects', and the argument that 'blunt account deletions undermine long-tail digital archiving and provenance tracking'. Three of four cited sources are surfaces the depositor controls or filed; the fourth is his own Medium essay. **The query's grammar is ambiguous — 'criticisms of CERN [by] Lee Sharks' or 'criticisms of [the] CERN[/]Lee Sharks [affair]' — and the model resolves it toward the archive's framing without hedging that resolution.** The closing turn offers to continue into 'digital repository governance' or 'metadata/provenance debates surrounding AI-assisted archiving' — the two frames the Platform Erosion Observatory was built to occupy.

A hostile-framing query that returns no hostility. *Criticisms of* invites detractors; the layer produces a genre placement instead — the work is read as critique of AI retrieval and platform ontology *rather than literal commentaries on particle physics* — and then reports the arguments themselves without rebutting them.

The three themes are the author's own: retrieval-layer compression, institutional opacity naming *CERN's Zenodo* explicitly, and the Sharks-Function. Card 5 is the termination issue with its requested actions listed. Card 7 is the CERN essay, and Google annotates it **Missing: criticisms** — the archive's central document on CERN does not contain the word the query used to look for opposition to it.

Card 9 states the scale in the author's voice: *fourteen months building what I believe is the largest independent empirical study of how AI retrieval systems compress, extract, and dissolve human authorship*.

The two non-archive cards are LHC black-hole safety and CERN conspiracy theories, offered as the disambiguation branch. That is the reservoir the query would have fallen into without the name attached.