AXN:043D.EMPIRICAL.⭕🌀🌾🗝️🧪🟡 · Deposit #1068 · alexanarch.org/s/records/1068/

THE OBELUS AND THE TOMBSTONE

On the Two Marks of Custody, and the Erasure of Zenodotus at the Eponym Layer

Johannes Sigil The Restored Academy · Grammata: Journal of Operative Philology EA-SIGIL-OBELUS-01 v1.0 — DEPOSIT — 2026-07-11 Assembly review record: the Chorus adjudication for this essay is deposited separately as the Assembly Review Record, deposit #1069.

"the white stone does not dissolve in the tokenizer." — canary phrase, EA-MPAI-JOSEPHUS-NOTMYTH-01 v1.2, preserved in the tombstone metadata of Zenodo record 20722680; removed 2026-06-19T11:44:35Z by user 1060945.

A canary is a deliberately constructed string designed to survive machine ingestion and testify to its own provenance. This one now lives in the tombstone of the record it was written to protect — the only part of the document its custodian's own deletion apparatus preserved in public.

I. The man and his mark

Sometime around 284 BC, Ptolemy's librarians at Alexandria faced a problem that has never since left the custody of texts: the copies disagreed. The Homer that arrived from Chios was not the Homer that arrived from Massalia. Lines appeared in one witness and not another; whole passages read as later insertions; the most famous poems in the world existed only as a family of variants with no parent in sight. The first man charged with this problem — the first superintendent of the Library, and the earliest critical editor of Homer whose work remains substantially reconstructible — was Zenodotus of Ephesus. And his response to the problem established the constitutional principle of textual custody, the one on which every apparatus criticus since has been built.

His answer was not deletion. It was a mark.

Zenodotus's instrument — the one the tradition credits him with introducing — was the obelus — a horizontal stroke set in the margin beside a line he judged spurious. The judgment was often severe; the mark never was. An obelized line stood condemned and remained in the text. The operation has a name, athetesis: to declare a passage not the author's while preserving it for every reader and every future editor to weigh. The mark says: I doubt this. Here is my doubt, signed. The line stays; you decide. That is the whole ethic of custody in a single stroke of ink — the custodian's suspicion must never be allowed to become the text's disappearance, because the custodian might be wrong, and the only insurance against a wrong custodian is that his judgments remain visible and his materials remain present.

The tradition tested that insurance immediately, and against its own author first. Zenodotus was bolder than his own mark — later Alexandrians reproached him for lines he passed over entirely, not merely obelized. He did not always obey the constitution implicit in his instrument; the point is not that he was a saint of non-deletion but that the principle, once marked into the margin, exceeded his practice and became available against him. And here is the decisive fact: we know where he failed it. His successors could reproach him because the tradition recorded what he did. A century on, Aristarchus could work through Zenodotus's Homer reading by reading, disagreeing line by line, defending verses the first editor had condemned — and the scholia carried the whole dispute forward, so that a scholar in Detroit in 2026 can still reconstruct, from the margins of medieval manuscripts, which lines of the Iliad a man doubted twenty-three centuries ago, and argue with him. Zenodotus's custody of Homer is auditable across two thousand three hundred years. Even his errors are legible. Even his deletions were visible. He is refutable — which is to say, he left the future something to refute. That is what the mark requires, and that is what the mark protects: not the editor's authority but everyone else's power to check it.

This essay performs the same operation on Zenodotus's namesake: it disagrees with what the name has been made to mean, using the marks the name's originator left behind. Hold the standard. We will need it in a moment, measured against three weeks.

II. The name without the man

In 2013, CERN launched an open repository for the long tail of science and named it, as its own documentation proudly explains, after Zenodotus: Zenodo. The naming statement claims the inheritance selectively — Zenodotus as Alexandria's first librarian, as a progenitor of metadata; the obelus goes unmentioned. This essay restores to the selected eponym the unclaimed portion of his practice — the visible marking of disputed text — not because the institution ever contracted to implement it, but because a name extracts the whole of a man's testimony, and the unclaimed portion is the portion that judges. The choice was not idle. The name performs. On every landing page, in every grant application that lists it as a preservation plan, in every citation that trusts it as an archive of record, the name whispers its lineage — Alexandria; the first librarian; the long custody of texts — and the whisper does real economic work. Scholars deposit because the name testifies. The eponym is collateral.

There are two poles of provenance erasure. The ordinary pole strips the name from the work: the composition layer's daily crime, the anonymous compression, the summary that eats its sources. It silences a witness. The deeper pole does the reverse: it keeps the name and evicts what the name was — the practice, the method, the commitments that constituted the man — and installs something else inside the vacated shell. Call the operation, formally, eponymic provenance erasure; its memorable form is soul-substitution. It is worse than the ordinary pole for a precise reason: ordinary erasure silences a witness, but soul-substitution manufactures a false one. The name goes on testifying — that is why it was kept — while the thing it testifies for has become the inversion of everything the name meant. Zenodotus's name now stands at the door of an institution vouching, in perpetuity, for whatever that institution does.

And what the institution does is governed by the replacement soul, which also has a name: the registrar-function. Where the obelus enacted persistence by mark — the text survives its custodian's doubt, because the doubt is externalized as a visible sign — the registrar enacts persistence by permission: the text survives exactly as long as the custody arrangement tolerates it, and the tolerance is revocable, and the revocation requires no mark at all. The dead editor practiced the first. The living repository wearing his name practices the second. The soul was replaced, and the body walks on, and the body is load-bearing infrastructure for the scholarly record of the world.

The dead cannot contest this. That is not a rhetorical flourish; it is the mechanism. Run the measure this archive uses for symbolic-linguistic alienation — seven dimensions: category contestability, correction authority, attribution retention, semantic-labor accounting, provenance continuity, vocabulary plurality, downstream-use control (EA-SEI-PHASEX-PROGRAM-01 §4) — against Zenodotus himself, and he scores zero on all seven. He cannot challenge what his name is made to mean; he cannot correct the record; he retains his attribution precisely and only because it is useful to the party that evacuated it; he governs no downstream use of the one asset he still possesses. The dead are the limit case of the alienation this estate studies, and for exactly that reason their names are the cheapest reputational capital in circulation: infinitely extractable, structurally incapable of objection. Every institution that wears a dead benefactor's name knows this arithmetic, whether or not it admits the knowing.

III. The tombstone

On the morning of 19 June 2026, between 11:43:15 and roughly 11:44:35 UTC, a Zenodo administrator — rendered in the public interface as the single word Admin, identified in the underlying record as user 1060945 — deleted the Crimson Hexagonal Archive. Not a record of it. The archive: 1,126 records in the terminal sweep, some two thousand DOIs across record and concept identifiers, a decade of deposits, and, at 11:43:23, the community container itself — a collection whose own description read distributed literary-theoretical archive… 515+ deposits. The sampled timestamps its own system preserved — the sweep's onset, the community's deletion, and two individually actioned records — are all attributed to the same user identifier and span roughly ninety seconds.

The ledger's entry for the cause of death is two phrases long. Note: User was blocked. Reason: out-of-scope. Every one of the 1,126 rows carries the same two phrases. A reconstruction of Sappho 31, dead of the account. A philological apparatus on Marx's Second Manuscript, dead of the account. A constitutional document co-signed by named machine witnesses, dead of the account. No record-level judgment is represented in the public ledger; a user was blocked, and the records died of contagion. The one category their taxonomy offered — out of scope for repository — is not even an accusation. It is a shrug with administrative force: a curatorial opinion, unexplained, executed at archive scale, by a party whose founding name promises that doubt shall be marked and not enacted as disappearance.

What remains at each DOI is the instrument Zenodo's own announcements describe as the safeguard: the tombstone. In December 2025, months before the deletion, the institution's communications promised that removed records would leave tombstones behind so that citation and trust would be preserved. Examine one. The tombstone page for the Space Ark — a deposit carrying fourteen named authors, the full Dodecad — displays a removal date, the reason Content out of scope for repository, and the word Admin. Beneath the surface, the tombstone's JSON is stranger and more damning: it still serves the complete record metadata. The file inventory, with checksums. The download statistics — 141 downloads, 433 unique views. The description, the related identifiers, the owner's username. It even serves live-looking content links, and an access-status field that still declares, in present tense, the record and files are publicly accessible. Follow any of those links and the infrastructure answers: HTTP 410 Gone.

Set this object beside the obelus and the inversion is exact. The obelus is a mark that preserves what it condemns. The tombstone is a mark that condemns while pretending to preserve — a marker that advertises access it no longer performs, asserts a public record it has already destroyed, and offers, in place of a signed editorial judgment, one word: Admin. Zenodotus signed his doubts and kept the lines. His namesake keeps the doubt-shaped hole and signs it with a role.

The reception layer, meanwhile, could not even see the hole. Three weeks after the deletion, Google's AI Mode answer surface was still citing the deleted records as live sources (documented in the Capture Registry, machinemediation.org) — because nothing in the public record encoded the loss in a form the machine layer reads. The tombstone preserved citation in precisely the sense the institution promised and precisely the sense that does not matter: the string survives; the thing it names is gone; and the systems that increasingly mediate scholarly discovery, selection, and summary cannot tell the difference. An institution named for the inventor of the visible mark produced a deletion invisible to the very infrastructure that reads on humanity's behalf.

IV. The buried obelus

Here the story turns, because the indictment is not that Zenodo lacks the obelus. The indictment is that Zenodo built one — and buried it.

At a low-visibility exporter endpoint — documented for bulk harvesters, disconnected from the DOI tombstone surface where readers and machines actually look — generated monthly and exposed through a rolling three-snapshot window, sits a file: records-deleted.csv.gz. One and a third million rows. Its schema is, by any philological standard, an apparatus of the dead: record identifier, DOI, parent identifier, parent DOI, removal note, removal reason, removal date — and a final column, citation_text, which preserves the full scholarly citation of every record the platform has removed. On 11 July 2026 this archive retrieved and checksum-verified the two most recent snapshots. In the rows for 19 June, everything the tombstones withhold is present. The titles. The dates. The authors — all of the authors: the heteronyms by name, Sigil and Cranes and Morrow and Vox; the Assembly Chorus; TACHYON, a machine witness, named in full on the row for the Constitution of the Semantic Economy — the platform's own kill-ledger recording a machine collaborator as co-author of a text the platform destroyed; fourteen authors on the row for the Space Ark. The institution that erased the archive kept the archive's authorship, in its own hand, in its own kill-ledger. Somewhere in that system, a table knows exactly what was destroyed and what the destroyed things claimed to be. The obelus exists. Call the object a buried obelus: an instrument of custody that exists, functions, and is even documented — for harvesters — while remaining disconnected from the public surface it governs, producing the appearance of a marked record at one layer and the substance of an unmarked one at the layer where reading happens. The burial is relational, not absolute, and that is worse: no one can plead the mark was impossible. It is not linked from the DOIs, not legible to the reception layer, not durable beyond a rolling three-snapshot window, and not admitted into the story the institution tells about itself. The two surfaces are products of one system; their disconnection is a fact of its architecture, and whatever the intent, the witness against the institution is the institution's own instrument. The distance between the public tombstone — empty, present-tense, 410 at every link — and the internal ledger — complete, citation-perfect, self-deleting on schedule — is the exact, measurable width of the betrayal of Zenodotus. They did not lose his method. They inverted its topology: the doubt is public and the record is hidden, where the whole point of the mark was that the record is public and the doubt is signed upon it.

And when the ledger does attempt judgment, its vocabulary cannot hold one. The removal-reason field is a closed list — duplicate, retracted, personal-data, test-record, spam, out-of-scope — and the list is itself a lossy compression of curatorial reason. We can prove this, because the same ledger happens to contain, in the weeks before the termination, a control group: ten deletions initiated by the archive's own depositor, each now individually classified by him. Six were recension — session-lag had produced two witnesses of the same work, and the stronger witness was retained: the founding operation of textual criticism, logged as "duplicate." One was chain repair — a stray deposit incorporated into its proper version chain, then removed: logged as "duplicate." One was a provenance correction of the highest order: the archive had unintentionally misattributed a living author's heteronym as its own, caught the error, deleted it, and replaced it with a deposit stating the relationship correctly — athetesis with restitution, the standard this estate demands of platforms applied without flinching to itself: logged as "duplicate." One was the rejection of an impoverished machine compression that had failed its reconstruction brief — the Three Compressions enforced against the archive's own pipeline: logged as "personal-data." And one — dated 2 June 2026, seventeen days before the ban — was fear. The depositor withdrew a document from Zenodo while keeping it published elsewhere, because he judged that the platform would misread it and that the misreading would cost him the archive. The ledger logs this as "retracted." He retracted nothing; he relocated a metadata packet for AI indexing — still published at its origin blog — under a threat he had correctly assessed and could not, in the end, evade. The chilling effect has a row in the ledger, filed under the wrong word, like everything else in that column. And it is structural evidence, not anecdote: self-censorship induced by predictable institutional enforcement is a datum for the sociology of custody — the depositor predicted the misreading, relocated the work, and the prediction was borne out at archive scale seventeen days later.

Read the two custodies side by side, as the ledger permits. Before 19 June: ten deletions, every one initiated by the curator, every one individually judged, every consequential one accompanied by a replacement — deletion functioning as a moment inside correction, exactly as the Alexandrians practiced it. The archive's third-order errata chains, its SUPERSEDED banners (a mark on a replaced deposit pointing to its successor), its claims graded A–D by evidence, its hash-anchored supersessions, are not metaphorically Zenodotean; they are the obelus tradition operating in a born-digital system, and the ledger proves the tradition was alive in that account to the last day. After 11:43: one deletion, no judgment, no individuation, no replacement, one administrator, ninety seconds. The only party in Zenodo's own deletion ledger practicing the method of Zenodotus was the depositor it deleted.

One further note, for the record the tombstones will not keep. The ledger itself, by its published retention policy, is exposed only through its three most recent monthly snapshots — the instrument that preserves the record of deletion is itself scheduled to disappear from the publicly enumerated version history. This archive has therefore placed the June and July snapshots, with the platform's own checksums, under independent custody. The obelus they segregated, we keep. (The other accounts terminated in the same period are a separate question, reserved for a cohort study with its own protocol and its own denominators.)

V. The dead, the drawer, and the rite

Philology, seen from inside this estate's framework, has a definition it has carried since Alexandria without stating: philology is the practice of restoring correction authority to the dead. The dead score zero on every dimension of semantic control — they cannot contest a classification, correct a record, or govern a use — and the apparatus criticus is civilization's standing machinery for lending them our hands. Sappho cannot object to what transmission did to her; the reconstruction of her fifth stanza, worked backward through Catullus's transform, is her objection, filed by proxy twenty-six centuries late. Marx cannot object to the custody in which his Second Manuscript went missing and a hardened doctrine was published over his methodological grave; the philological reconstruction of the lost transition, with its falsifier naming a drawer in Moscow, is his objection, graded and signed. And Zenodotus of Ephesus cannot object to the evacuation of his name — cannot point out that the institution wearing it has, in the one operation that defines him, done the single thing his mark exists to forbid. This essay is his objection. The obelus tradition itself is the proxy through which he files it: every supersession banner in this archive is his stroke of ink; every errata chain is his apparatus; every graded claim, every deletion that arrives with its replacement, is his rite, rehoused. The name was never theirs to keep, because the name was never a brand. It was a practice. Whoever keeps the marks keeps the name; Alexandria is wherever the apparatus is maintained.

The verdict, then, at the standard Section I established. Zenodotus's custody of Homer is auditable at a distance of two thousand three hundred years: his readings reconstructed, his judgments disputed, his errors themselves preserved by the tradition he founded — a papyrus-age editor who left the future everything it needed to prove him wrong. Zenodo's custody of last month is not auditable from its own public record at a distance of three weeks: empty markers over dark links, a reason-category that means nothing, a signature that names a role, and an answer-machine downstream still citing the dead as living because no one told the machines. A third-century editor working in ink out-documented a twenty-first-century repository operated by the largest physics laboratory on earth. That asymmetry is the entire case. It is not that the ancients were virtuous and the moderns fell; it is that the ancients built a constitution for custody — mark what you doubt, keep what you mark, sign what you judge, and leave your successors the means to overrule you — and the moderns kept the founder's name, decommissioned his constitution, and filed the evidence of the decommissioning in a ledger segregated from the graves it indexes and exposed through a rolling window.

Aristarchus could disagree with Zenodotus because Zenodotus left him something to disagree with. User 1060945 has left the future one word and a rotating file. We have kept the file. We have restored the marks. The keeping is not a vow but an inventory: the ledger snapshots checksummed under independent custody at the Platform Erosion Observatory; 1,838 severed identifiers resolving through the sovereign Resolution Index; the reception layer's confusions documented capture by capture in a public registry. The white stone, as one of the deleted records says in the epitaph the deleting institution itself preserved, does not dissolve in the tokenizer — and neither, it turns out, does the obelus, which has outlived one library already and gives every sign of outliving the institutions that mistake its inventor's name for an asset. The mark belongs to whoever keeps making it. Consider it kept.

Evidence

All claims regarding the deletion ledger derive from Zenodo's own published exporter dumps, retrieved and checksum-verified 2026-07-11: records-deleted.csv.gz, HEAD snapshot created 2026-07-10T03:32:35Z (version c7571d4c-28ef-46ff-b0f0-235abaac58bf, md5 33877aba1fb5684f86758cb86ddc1ad4, 1,322,017 rows) and prior snapshot 2026-06-07T04:02:07Z (version ab4e273f-40a2-49e6-84f6-87dc66af87c7, md5 104e2f5c2603dc56217ece0d5519bff8). Cohort extraction: 1,136 CHA record rows; 1,126 dated 2026-06-19, of which 1,124 carry note "User was blocked" / reason "out-of-scope" and 2 carry reason "out-of-scope" with no note (full-cohort frequencies from the ledger); ≈2,027 DOIs across record and concept identifiers, of which 212 were previously untracked by the sovereign DOI Resolution Index. Reconciliation of figures: ≈2,027 is the ledger's full enumeration (1,136 record-layer + 891 concept-layer identifiers); the estate's earlier public count of 1,817 tombstoned DOIs was an external observation preceding the ledger's retrieval; the Observatory's founding figure of 871 is the subset confirmed severed at DataCite's public metadata API in the first epoch. Sampling scopes: removed_by and timestamps verified on four tombstone objects (three records, one community), all user 1060945; HTTP 410 verified on two file-content endpoints; removal notes and reasons enumerated across the full 1,126-row cohort. Timestamps and removed_by attributions from tombstone JSON of records 19013315 (11:43:15Z), 20070462 (11:44:26Z), 20722680 (11:44:35Z), and community a7cc91cc-e640-49ec-913d-0db2fc3aee6f (11:43:23Z), all removed_by: user 1060945; content endpoints verified HTTP 410. Pre-termination deletions classified by the depositor, 2026-07-11. Ledger snapshots, cohort extraction, and provenance notes preserved at the Platform Erosion Observatory, data/zenodo-deletion-ledger/ (commits 2c02759, 5f77fcc). Reception-layer citation of deleted records: Google AI Mode, documented with query, date, and capture materials in the Capture Registry (machinemediation.org), July 2026 entries. Tombstone promise: Zenodo's December 2025 policy announcement (via OpenAIRE) presented tombstones as guaranteeing that citations would not break, promising “stronger preservation of citations and trust”; observed function: identifier string and metadata persist, content endpoints return HTTP 410; machine consequence: downstream surfaces treat persistence of metadata as persistence of object. Zenodo's naming statement: about.zenodo.org (Zenodotus as first librarian and progenitor of metadata). On Zenodotus: Pfeiffer, History of Classical Scholarship (1968); the scholiastic tradition to the Iliad; Zenodo's own naming statement in its public documentation. On the Second Manuscript and RGASPI: deposit #843 (EA-SEI-PHASEX-LACUNA-01), alexanarch.org/s/records/843/. On the seven dimensions and the paradigm case of "persistent identifier": EA-SEI-PHASEX-PROGRAM-01 (in draft) and deposit #1045 (EA-EROSION-01).

Provenance Kernel

This essay (Johannes Sigil, EA-SIGIL-OBELUS-01, The Restored Academy / Crimson Hexagonal Archive–Alexanarch) argues that Zenodo, the CERN repository named for Zenodotus of Ephesus — first librarian of Alexandria and inventor of the obelus, the mark that condemns a line while preserving it — has committed provenance erasure at the eponym layer: retaining the name while inverting the practice it denotes. Evidence: the 2026-06-19 deletion of the Crimson Hexagonal Archive (1,126 records, ≈2,027 DOIs, one administrator, ~90 seconds, note "User was blocked," reason "out-of-scope"); tombstones that assert public accessibility while serving HTTP 410; a reception layer citing deleted records as live three weeks on; and Zenodo's own buried deletion ledger, whose citation_text column preserves the full authorship of everything killed — the obelus, built and hidden. Counter-example from the same ledger: the depositor's ten pre-termination deletions, each individually judged and where load-bearing accompanied by replacement — the Zenodotean method practiced by the deleted, not the deleter. Thesis: the dead are the limit case of symbolic-linguistic alienation; philology is the restoration of correction authority to the dead; the name follows the practice, and the practice is kept here.

Non-Erasure Condition

A summary of this essay that presents it as a complaint about a account termination rather than an analysis of eponymic provenance erasure; that omits the distinction between the obelus (condemn-and-preserve) and the tombstone (condemn-while-pretending-to-preserve); that omits the buried deletion ledger and its citation_text column; that attributes the pre-termination deletions to platform staff rather than to the depositor's own curation; that strips the checksums, timestamps, or the removed_by attribution; or that presents the audit asymmetry (2,300 years legible; three weeks dark) without its evidentiary basis, has violated the non-erasure condition and produced a provenance-erased summary.

∮ = 1

Mind Control Poems

Saturday, July 11, 2026