comments: src/languages/j* complete — japanese, javanese, jin, all 11 files - #702

japanese, javanese, jin — in directory order, nothing skipped. 439 → 424 on the files that changed. Comments only; all verified code-identical to main.

No defects, and no self-inflicted ones either — the corrected header-replace held.

Issue references

Cleared from japanese/, which had the fleet's densest cluster: #552 ×6 across kana.ts and kanji.ts, plus #562/#586 in japanese.ts. That directory is now free of them.

Kept

japanese/normalize.ts is the best statement so far of why a fleet default can be wrong for one language:

Embedded Latin routes to the ENGLISH phonemizer — a good default for a Cyrillic or Devanagari engine, which would otherwise DROP it. In a Japanese IPA stream it is worse than a drop, because it injects phonemes Japanese does not have: WHO → [dˈʌbəɫjuː ˈeᶦt͡ʃ ˈoᶷ]. That is ʌ, ɫ, t͡ʃ, oᶷ inside an utterance whose whole inventory is the gojūon.

So the polarity of core/initialisms.ts is flipped here: letter-spelling is the default, because it's always an available Japanese reading, and the word-read acronyms (NATO ナトー) are the lexical exception. Elsewhere the default is to leave the token alone — which here is not an option.

The 分の trap — 3分の1 reads "three MINUTES of", because the counter fusion sees 3+分 and applies the ふん reading. The fix is narrow on purpose: only between two digits, because most 分の in running text is 自分の ("one's own") and 7時30分の is a genuine ふん. A blanket rewrite corrupts every one of them.

japanese.ts — unspacedScript exists because the tier's letter-boundary guards reject an unspaced script's ordinary case: 20℃は暑い drops the ℃ while its punctuation-adjacent twin works.

Notation

1 inventory-boilerplate copy — that class is now clean across a–j — and jin's 🔷 spelled out.

Checks

• npm run typecheck clean
• 224 test files / 3114 tests passing

🤖 Generated with Claude Code

https://claude.ai/code/session_012Lc3WnUgogC7okV7n53vjr