A GRIEVANCE ISSUE DOUBLING AS AN INFRASTRUCTURE ANNOUNCEMENT. Card 2 is the GitHub issue reporting the account block, and the snippet the layer selected is not the complaint but the offer: "alexanarch.org — sovereign successor archive, 879 deposits, content-derived AXN identifiers, distributed custody. OPEN TO DEPOSITORS NEEDING POST-ZENODO INFRASTRUCTURE."

The deposit count is another frozen census — 879 against a live 1,457 — and this one is dated 18 February, which predates both the termination and the successor archive. The issue has evidently been edited since, and the layer serves the edited state under the original date.

Card 1 states the design principle in one line: "Declared provenance is a FEATURE OF ITS ARCHITECTURE, NOT A COMPLIANCE BURDEN." Card 3 names the purpose the whole apparatus points at — corpora built "for FUTURE TRAINING PIPELINE INGESTION at scale" — retrieved by a system that ingests.

AI Mode retrieves alexanarch cleanly as 'an emerging decentralized, post-Zenodo digital archive network (often referenced as alexanarch.org)' — 'distributed custody and open-access data deposits for users needing alternative preservation infrastructure, alongside discussions regarding provenance tracking in training data pipelines.' Three sources cited: (1) Medium — Lee Sharks, 'Provenance Debt' (July 1, 2026): 'Alexanarch is one of the counter-friction operations. Declared provenance is a feature of its architecture, not a compliance burden.' (2) GitHub — Zenodo issue #2596, 'Account blocked and records removed — request manual review' (Feb 18, 2026): 'alexanarch.org — sovereign successor archive, 879 deposits, content-derived AXN identifiers, distributed custody. Open to depositors needing post-Zenodo infrastructure.' (3) Academia.edu — 'Provenance Debt and the Extraction Economy of Unmarked Augmentation' (July 2, 2026): 'The Alexanarch archive is named as one of the counter-friction operations building provenance-designated corpora for future training pipeline ingestion at scale.' Follow-up offered: 'If you are looking to browse the web for specific records or want to understand how this platform's open-access data differs from traditional repositories, let me know.' The archive's own identity as sovereign-successor-to-Zenodo is now the composition layer's first-frame characterization — retrieval basin stabilized. 879-deposit figure is stale (current: 1078 post-#1078 mint of this same date).

THE ARCHIVE’S SELF-DESCRIPTION IS NOW THE DESCRIPTION. Four of four sources are archive or authored, and the composed answer reproduces the founding statement — sovereign infrastructure, content-derived AXN addresses, distributed custody, tombstone permanence, "founded after 871 DOIs" — without qualification or attribution to a claimant. Card 2 is worth its own note: it describes a deposit as "the FIRST MACHINE-AUTHORED DEPOSIT in Alexanarch", so the layer is composing an account of machine authorship inside an archive that admits it, cited from the archive’s own wiki.

THE FIGURES ARRIVE WITH THEIR AUDIT TRAIL. Card 2 carries the count AND how to check it: "of 1,817 affected DOIs, 871 (47.9%) now return HTTP 404 from DataCite’s public metadata API at api.datacite.org/dois/{doi} — independently verifiable". Card 3 carries the refinement that matters legally: "851 of the 871 severed DOIs are UNIQUE WORKS: they share no title with any preserved DOI and are not an alternate version of a preserved record" — the disposal of the obvious rebuttal, retrieved alongside the claim.

Card 1 is a genuine prior occupant of the name: a SoundCloud musician recording as Alexanarch, with tracks dated three and four years ago. This corrects an assumption in an earlier reading of «alexanarch sappho room», where I wrote that "alexanarch" is "a coined name with no prior occupant". It has one. The collision is quieter than "crimson hexagon" — one artist rather than a market-analytics firm — but it exists and predates the archive’s use.

CORRECTION, MANUS 2026-08-12: I wrote that "alexanarch" HAS A PRIOR OCCUPANT and corrected an earlier reading on that basis. Both statements were too strong. The SoundCloud artist is ALEX ANARCH — two words, a different name, apparently inactive. The layer matched across the space.

So the original reading stands after all: "alexanarch" as a single token has no prior occupant, which is why it holds a total archive basin where "crimson hexagon" does not. What this capture adds is a distinct collision mechanism — WHITESPACE-INSENSITIVE MATCHING — which belongs in the register beside root-sharing, acronym contest and exact-name contest, and which is weaker than all three: it establishes nothing about prior use.

The five principles are returned as the archive states them — Obelus, Substrate Disclosure, No Classifier Training, Sovereign Identity, Network Preservation — with AXN correctly described as byte-anchored and derived from the text hash. This is the archive's own governance vocabulary reproduced without dilution. But the institution is named only through its code host: all three cards resolve to GitHub or SoundCloud, and alexanarch.org appears nowhere. The layer knows the doctrine and does not know the address. The SoundCloud card is a bare name collision seated as though it were a source about the archive.

Seven hours after the 16:44 capture at this address, the same string returns a different answer and a better one. That capture cited GitHub and SoundCloud and never named a person; this one opens with *founded on June 19, 2026, by researcher Lee Sharks*, states the termination as the founding cause, and cites alexanarch.org three times including the Principles page and the token-level search surface.

The Tombstone Guarantee is stated in the archive's own terms — content *cannot be silently removed or vanish without an archived trace* — and the Obelus Principle card carries Zenodotus and the line that the mark followed from reading rather than pattern-matching.

Two variables changed between the sittings: authentication (signed out and incognito at 16:44, signed in here) and time. The corpus cannot separate them from a single pair.

**A CORRECTION TO THE FIRST DRAFT OF THIS READING.** I recorded that Google had disabled AI Overview in incognito and that the authentication-controlled pair was therefore gone as a method. That is wrong, and the operator has observed the behaviour many times: **the incognito Overview is a switch Google toggles on and off periodically.** It is intermittent, not withdrawn.

Which makes the availability itself measurable rather than an obstacle. This registry holds 96 observations that are both incognito AND Google AI Overview, and every one of them dates a moment when the switch was ON — 2026-06-27, 06-30, 07-13, 07-14, 07-15, 07-18, 07-25, 07-26, 07-31, 08-03, 08-04, 08-05, 08-06, 08-08, 08-09, 08-11 and 08-13. What the corpus cannot yet see is the OFF windows, because a capture that produced no Overview was never seated as an observation. NULL is not zero: an unattempted incognito query and a refused one are indistinguishable in this record, and only the operator's own attempts distinguish them.