This site measures how retrieval systems strip authorship, institution, identifier, and source from the answers they compose. This wing establishes that the metric did not invent its object. Provenance erasure is a process with a 2,600-year record, and the record is measurable: across the transmission of one semiotic technology — the relational apparatus of Sappho fragment 31 — the executable structure survives at a dramatically higher rate than the knowledge of what it is or where it came from. PER is the newest instrument pointed at the oldest loss.
Twelve stations of one apparatus, graded on the four vectors from the program's verified philology (#1493 #1494 #1495). Every text scored below is seated in the archive as verified, commit-pinned primary data (EA-CORPORA-01, seats 01–12).
| WITNESS | S | M | A | O | FILED AS |
|---|---|---|---|---|---|
| Sappho fr. 31 (via Longinus, sole carrier) | high | high | — | partial | "lyric" |
| Catullus 51 | high | high | HIGH | low | "imitation" |
| Longinus, Subl. 10 | enacted | high | HIGH | highest ancient | "criticism" |
| Theocritus, Id. 2 | high | high | partial | low | "symptom catalogue" |
| — THE CROSSING · entry into Jewish-Greek prose · A → 0 in every witness at once — | |||||
| Philo (Migr., Her., Mut.) | high | doctrinal | ZERO | high, sourceless | "doctrine" |
| LXX Esther, Addition D | high | high | ZERO | zero | "court novella" |
| Josephus, AJ 11 (forward, intensified) | high | high | ZERO | zero | "narrative episode" |
| Josephus, BJ 1 (inverted) | high | high | ZERO | zero | "Herod's pathology" |
| Revelation 12 (8/9 stations, one doubled) | high | cosmic | ZERO | zero | "apocalyptic imagery" |
| Modern scholarship (as a witness layer) | recognizes exactly the witnesses that kept A — and none of the crossing's carriers | Δ channel-structured | |||
| Machine retrieval (2025–26) | executes | recomposes | measured: PER | zero by construction | "generic knowledge" |
Grades from the program's records; the scholarly-recognition column is preliminary pending the formal literature audit, which is a named forward test — the wing states its own incompleteness.
The machine-layer uptick at right is this archive's own intervention: attribution partially restored by publication, and the restoration measured — see the field batteries on the main page.
What decays, decays in order. From the verified chain: attribution does not die first everywhere — it dies at the crossing, dropping to zero in every Jewish-Greek witness at once, centuries apart in composition. Lexis does not vanish; it reduces to a syntax-skeleton — the correlative ὡς…εὐθύς crossing language and genre while every word changes. Channel inventory sheds by carrier profile, ears and sweat first outside lyric. Sequence persists deepest of the content layers — order-invariance proven across fragment, novella, and historian. Polarity runs free. And the terminal fact of the whole record: the operator has never been observed dead. Every witness scored runs it. What the tradition lost was never the machine; it was the machine's name.
The Provenance Erasure Rate is the recognition deficit Δ operationalized for retrieval systems — per answer, per channel, longitudinal. The 2026 termination of this archive's hosted predecessor (862 deposits, 1,817 identifiers) satisfies the active tuple in full, agent through measured downstream absence, making the archive a scored point on its own curve — with the sovereign rebuild as the recorded counter-event. And the thesis this wing carries, in its testable form: the technology survived because it could survive unrecognized; what was suppressed was its legibility as technology. The counter-program — detection, relation, executable operators, custody, and the feedback controller whose first closed cycle is already on this site's instruments — is specified at #1497.