The dating pendings close — three gaps, three different causes
7 August 2026 · Nabu news
The coverage survey said three sources held dates the library wasn’t reading: ORACC (~82,000 documents), Rundata (~24,000), and eBL (~23,000). All three closed this wave — and each gap had a cause the survey didn’t predict, which is the argument for scouting before building.
ORACC’s was one glob depth. The catalogue files that carry period labels live at varying depths under each project, and the biggest project of all — the Ur III administrative corpus, eighty thousand tablets — unpacks its catalogue three levels down where the extractor walked two. The fix is a recursive walk and a real triply-nested fixture pinning it. The same catalogue read now emits each tablet’s period as an ordinary document facet, which means the Akkadian and Sumerian stage rules — which have mapped period labels to historical stages since the lect layer landed — reach ORACC’s own labels too.
Rundata’s was inheritance. The undated documents were never undatable: they are the translation and normalization lanes of stones whose base inscription parsed its scholarly dating years ago. The lanes now inherit their stone’s band through the shared signum — a rendering of a dated artifact is dated by the artifact, the same reasoning the ORACC English witnesses have always used.
eBL’s was vocabulary. The survey flagged forty Seleucid-era day/month objects needing era conversion; the real pending was twenty-three thousand documents carrying plain period labels (“Neo-Assyrian”, “Sargonic”) that no dating lane read. They band now — and the Seleucid forty band coarsely under their period, with exact era conversion recorded as deliberately out of scope.
Under all three sits one new ruled artifact:
config/period_bands.yml, a single period→band table whose
conventions are cited row by row — CDLI’s own parenthetical dates
(“Ur III (ca. 2100-2000 BC)”) where the field states them, middle
chronology elsewhere, and honest refusals for labels like “Uncertain”
and for “Standard Babylonian”, which is a register, not a period, and
would mint an invented date. The timeline extractor’s private copy of
that knowledge retired in the table’s favor: one answer to “when is
Ur III?”, shared by every consumer.