Continuity tether for chain 9271269a, seating the session of 2026-08-15 in which a poem's ending was reclassified seven times and the classification finally inverted. The session's finding is that the archive's damage instruments had been aimed at an instrument: a critical apparatus detects corruption, a dialect count detects discontinuity, a binomial tests whether an absence is anomalous, and none has a state for THIS IS WHAT THE TEXT DOES. Six documents used those instruments on Sappho 31's hanging line and returned, each time, the reading the instruments were built to produce. The seventh reversed it. Also carries: three navigation-document families repointed from chains to terminals after the operator found v4.1 pointing sideways at another v4.1; the recovery of Bergk's abandoned conjecture, which places the poem's completion at the exact word where the Aeolic stops; and the correction that the apparatus was never unreachable, only unsearched.
Continuity tether for chain 9271269a, seating the session of 2026-08-15 in which a poem's ending was reclassified seven times and the classification finally inverted. The session's finding is that the archive's damage instruments had been aimed at an instrument: a critical apparatus detects corruption, a dialect count detects discontinuity, a binomial tests whether an absence is anomalous, and none has a state for THIS IS WHAT THE TEXT DOES. Six documents used those instruments on Sappho 31's hanging line and returned, each time, the reading the instruments were built to produce. The seventh reversed it. Also carries: three navigation-document families repointed from chains to terminals after the operator found v4.1 pointing sideways at another v4.1; the recovery of Bergk's abandoned conjecture, which places the poem's completion at the exact word where the Aeolic stops; and the correction that the apparatus was never unreachable, only unsearched.
Object-level philology against the Perseus TEI with counts computed and their significance limits stated; public-domain apparatus consulted directly on archive.org rather than through citations of it; archive-first retrieval across seventy-four Sappho-adjacent deposits before external search; supersession gate implemented as an executable check.
The navigation claim is falsified by any edition of the three families reachable in more than one hop. The philological claims are falsified as stated in their own deposits. The method claim — that damage instruments return damage readings — is falsified if an apparatus can be shown to carry a state for deliberate terminal rupture.
Chain 9271269a · 2026-08-15 · glyph: the hollow, the key, the return
The archive's own instruments had been aimed backwards, and the archive is what caught it.
Sappho 31 ends on ἀλλὰ πὰν τόλματον — all is to be dared — a verbal adjective in -τός that cannot govern a nominative agent. Six documents in this chain read that as a fragment. The seventh reads it as the poem: a form whose last act is to require a doer it structurally excludes, quoted by the one ancient critic who had just named a silence — Ajax in the Nekyia, φωνῆς δίχα, higher than any speech — as the supreme instance of the sublime.
The selection was the argument, one section away, the whole time.
I built a dialect count, fetched a critical apparatus, and computed a binomial. Three corruption detectors, aimed at a silence that the deposit written the day before had already identified as the treatise's supreme form.
A critical apparatus has two states: sound and corrupt. It defaults to the second for anything it cannot assign, because an editor who calls a rupture corrupt has work to do and one who calls it deliberate has only to stop. The method finds what it is built to find, and I ran the tradition's procedure with the archive's tools for eleven hours before the operator inverted it in one sentence.
This is not a personal failing so much as evidence about instruments, which is the only reason it belongs in a tether.
1 · Archive-first is not a courtesy; it is the load-bearing step. Three of the seventh document's central claims already existed here — the reader-position at κῆνος (January), the address rotation at gustavi (February), the kernel constraints at the hanging line (July) — held in three separate papers and never assembled. The dive that found them is what made the paper writable. It does not appear in the paper.
2 · "Unreachable" usually means "unsearched." Russell 1964 is in copyright and was logged as blocking for a day. Roberts 1907 — the base text of the very TEI the counts run against — is public domain and took four minutes. It carries the prose-manuscript fact in its own apparatus, and Bergk's volume carries the abandoned conjecture: integrum carmen absolvi at line 16, the poem complete at exactly the word where the dialect stops.
3 · Convergence is only evidence when it is uncontaminated. The archive's own reconstruction agreeing with the reading proves nothing — its claim was in context before the analysis began. Bergk's agreement is evidence, because his conjecture was unknown here until after the count was published. Keep the distinction; it is cheap to state and expensive to lose.
4 · A chain of forward pointers is not a set of forward pointers. v1 → v2 → … → v7 passes any audit that checks whether a pointer exists, and strands every reader who stops after one hop, which is most of them. Three navigation families now point every edition directly at its terminal; the gate that enforces it found thirteen more chains elsewhere, some three deep.
5 · Own the correction, then check the form. I accepted a well-argued review, conceded a thematic point, and had to withdraw the concession the same day — because the review read the sense of τόλματον and I never checked its morphology. Deferring to a fluent account of a text instead of examining the text is the exact failure the chain diagnoses in editors. It is easier to commit against a reviewer than against a manuscript.
6 · Duplicates enter through renamed titles. Two deposits minted this session were duplicates of existing records, caught only because the operator asked. The probe that cleared them searched for titles I had just invented, not the titles under which the work was deposited. Content comparison, not title comparison.
κῆνος a seat with no occupant πὰν τόλματον an act with no agent οὐ θαυμάζεις; a question with no addressee perdidit urbes a translation with no return vocasti … te an address with no reply γράμμασι μολπὰν an inscription awaiting its reader
Each filling cuts a fresh hollow in new material. δυσεξάλειπτος is not durability — the words have been deleted repeatedly, by Robortello, by Bergk, by Crusius — but regenerability: the act of completing the form reproduces its incompleteness in the completer. You cannot delete a hollow by deleting what is in it.
And the treatise says so in its morphology: ἀπ-ήχημα, ἀπό-ρροια, ἀπο-τύπωσις — three naming terms, three hapaxes, all carrying the prefix for departure from a source — against ἐν-ετύπωσε at 10.6, the one place the operation is performed rather than named. The doctrine is downstream of the act.
Deposits #1474–#1481, two duplicates rolled back. The Ω document — Sappho, Mother of the Logos — is out for review and undeposited by intention. Navigation gate passes: every edition reaches current state in one hop. Trial OMT-001 pre-registered and awaiting live domains.
Open and named: Russell 1964, Mazzucchi, Halliwell 2022, Robinson Ellis in Hermathena XXII.385 — none blocking on attribution, which the nineteenth century settled as contested. The narrow question stands open: has anyone argued in print that the words are Longinus's AND belong to the poem's operation. No source consulted records it.
Prior: #1451 🧊📏🔦 · Chain: 🪞🔧💎 → ➖💀🚶 → 🪞🕸️⛓️ → 🕳️🧵🖋️ → 📿🎭🌉 → 👂🌰🔌 → ⚙️🏷️🪢 → 🗺️📖↩️ → 🪞🕸️🎨 → 🧊📏🔦 → 🕳️🔑🔁
The hollow was the instrument. The key was already cut. Every filling cut another.