The character desk: the largest shelf gets its instruments
21 July 2026 · Nabu news
With Literary Chinese now the library’s largest language, Phase 37
builds the instruments the shelf deserves. nabu char 棄 answers one
character across every holding at once: radical and strokes, its
ideographic decomposition (Andrew West’s public-domain IDS data,
97,680 characters), component and variant links, readings from
Japanese kun and on through Mandarin and Korean — and then the column
no dictionary site offers: the Baxter-Sagart Old Chinese
reconstruction, the Qieyun-system reading, the character’s entries in
Heian-period Japanese lexicography, its senses in the Thesaurus
Linguae Sericae with their classical attestations, and its live corpus
counts. Structure search arrives beside it: by KangXi radical, stroke
range, or contained component — explicit modes, never silently mixed
into text search.
Retrieval learned the script’s realities. A traditional-simplified- variant fold derived from Unihan (6,050 pairs, with semantic variants deliberately refused — 直 and 值 are different words) folds 說, 説, and 说 to one search skeleton. Kanripo’s gaiji references were censused whole: only 36% of occurrences map faithfully to real codepoints, so exactly those render as glyphs and the rest appear as an honest placeholder with a counted footer — never a silently substituted character.
The Daozang joins the acquisition waves: 1,660 repositories of the Taoist canon, parsed through a witness-overlay extension that cites the reprint edition’s own page structure. The Thesaurus Linguae Sericae attestation crosswalk, re-censused against the completed Kanripo sync, resolves 80.2% of its 172,206 canon citations — a number that grows the day the Daozang lands, with no re-parsing, by design. And the phase closed a performance regression it found in its own instruments’ first full run: mega-document sources had slowed threefold under a batching change; the cause was isolated at fixture scale and the fix leaves them faster than before. Numbers as of 2026-07-21; shelf syncs owner-scheduled.