AI-generated conjecture · a wager, not a finding
← All conjectures · The cuneiform world
The dark residue of the best-indexed script
Status is derived only from the shepherd-authored triage/prediction data above -- community submissions and claims are a separate overlay and can never change it (see the participation panel below).
Claim (verbatim)
CDLI is the most complete catalogue any ancient script possesses - 421,501 artifacts. Yet its own descriptive fields carry a residue of the un-enriched: rows with no genre, no language, no assigned period. This is not a defect to apologize for; it is data. Cataloguing capacity is scarce and flows toward the interesting, so the vast administrative bulk receives a museum number and little else. Un-enrichment is therefore not random noise but a map of what the field found boring - the field's own dark stock, hiding in plain sight inside the world's best catalogue. The mechanism is triage under scarcity: enrichment effort is spent where scholarly interest is highest, leaving a structured shadow of unclassified rows that clusters exactly where interest is lowest.
Prediction clause (verbatim)
Prediction (measured on the in-house CDLI copy, 126,000 rows, about 30 percent of the catalogue): the genre-unclassified fraction is >= 8 percent (in-house exact: 10.4 percent, 13,098 rows), the language-undetermined fraction >= 12 percent (in-house: 14.3 percent, 17,990 rows), and - the structural clause - genre-blankness is not uniform: rows lacking an assigned period carry a genre-blank rate at least 3 times the genre-blank rate among period-assigned rows (primary clause: the period-null vs period-known genre-blank ratio >= 3). Disambiguation: genre-blank = empty genres field; period-null = empty period field. Coverage guard: the in-house copy is a partial harvest (pages 1-60 sequential plus an 80-page random sample), so the fraction is reported as the in-house rate, with the full cdli.earth API - which additionally exposes atf/translation presence, the deeper catalogued-versus-published gap - as the definitive re-run.
Kill-dataset (verbatim)
Kill: the CDLI genre/language/period fields - count blanks and cross-tabulate genre-blank by period-null. Runs THIS WEEK against the in-house CDLI copy, confirmable and extendable (to edition/translation presence) on the full cdli.earth API.
Nobody has run this test. The kill-data is named above. If you can run it — or you know the paper that already settles it — claim the kill or submit the prior scholarship. Kills and prior scholarship are credited here, by name, as they come in.
Provenance
Run: Fresh agent generation · model: claude-fable-5
Fresh blind generation by claude-fable-5, 2026-07-17, cuneiform instrument-anchored wave against CDLI (including the in-house catalogue copy), ORACC, BDTNS, and Archibab: every Kill names a specific open corpus and a countable operation (catalogue-incipit vs recovered census, series colophon N-of-M completeness arithmetic, un-enrichment residue, directional letter-asymmetry, dated-histogram discontinuity, lexical-vs-documentary attestation gap, intake-vs-holdings, curricular copy-frequency, godlist-vs-cult, chronology-vs-attestation, forgery-by-genre, provenance-null-by-genre, letter tablet-census, and genre/period composition shares). Disjoint by construction from the 30 cuneiform conjectures of wave w22 (grep-checked per key name and per instrument-operation). In-house CDLI copy = 126,000 rows (about 30 percent of CDLI's 421,501), harvested 2026-07-04; exact in-house aggregates cited in the runnable-this-week items: genre-blank 10.4 percent, language-undetermined 14.3 percent, provenience-null 11.9 percent, flagged fakes 160 (0.127 percent), Administrative 70.9 percent, Ur III 47.4 percent, Old-Babylonian:Middle-Babylonian dated ratio about 8. Six items resolve THIS WEEK against the in-house copy (ordinals 3, 5, 11-partial, 12, 13, 15); two carry an honest 'Kill (not yet built)' where the cross-instrument join is real in sources but not yet a dataset (ordinals 6, 7, 14).
Novelty / leakage triage
anticipated in the literature — this exact test has never been run
Two of the three clauses are the generator's own in-house measurements with thresholds pinned below the measured values (genre-blank 10.4 percent vs >= 8; language-undetermined 14.3 percent vs >= 12) - dead as evidence by self-measurement, exactly as the register admits. The designated primary clause, however - the period-null vs period-known genre-blank cross-tab ratio >= 3 - is not among the registered aggregates, and no published tabulation of CDLI field-blankness by period was located, so the one live clause is genuinely un-run.
- In-house CDLI catalogue copy (126,000 rows, about 30 percent of CDLI's 421,501 artifacts, harvested 2026-07-04), the wave's own registered aggregates
- CDLI, cdli.earth catalogue and API (the named full-catalogue re-run instrument)
Predictions
No prediction registered yet.
Weigh in
No community feedback yet.
New here? Create an account first
Create an account or sign in and your feedback is tied to you — you can track it, get replies, and claim this conjecture so others know you’re working on it. Prefer not to? Just leave your take below as a guest — only the name you type is shown.
Add your take
Posted immediately (spam is removed). Community feedback is never an adjudicated verdict and never changes this conjecture's triage label or status above.