The Japanese phase: Aozora Bunko and the honest-glyph ladder
21 July 2026 · Nabu news
Phase 38 answers a simple complaint honestly: a rare character shown as an empty box is honest but unhelpful. The new display ladder gives every not-yet-encoded character the most faithful rendering the evidence supports — the real codepoint where one exists, an ideographic description sequence (⿰氵丐) where only the composition is known, a visibly marked substitute ⌈…⌉ where the tradition offers a stand-in, and the placeholder box only as a true last resort, with a footer that always says what was substituted or left unresolved. Along the way the audit caught a real bug: 547 of the “faithful” glyphs shipped in Phase 37 were private-use codepoints — tofu on any machine without one specific font — and are now honestly demoted. Visible characters rise from 36% to 82% of rare-glyph occurrences, and none of them lie.
The phase’s second half opens the Japanese desk properly. An adapter for Aozora Bunko — the volunteer-built library of Japanese literature — covers its ~17,500 public-domain works: ruby (furigana) readings preserved as annotations rather than flattened into text, the base-text colophon carried as provenance, and Aozora’s own rare-character notation resolved through the JIS X 0213 standard mapping — while the fetch pulls the ~210 MB of actual text from a 23 GB upstream mirror by sparse checkout. A 173-pair old-form↔new-form kanji fold, derived from Unicode’s own name-kanji data and composed with the existing Chinese fold, lets 国 and 國 meet on one search skeleton, and the character card now cross-references both forms. Numbers as of 2026-07-21; the first Aozora sync is owner-scheduled.