comments: src/languages/g* complete — galician through gujarati, all 22 files

[christopherthompson81]

galician, gan, georgian, german, greek, guarani, gujarati — in directory order, nothing skipped. 926 → 823 comment lines. Comments only; all verified code-identical to main.

file	comment lines
greek/normalize.ts	282 → 259
gujarati/normalize.ts	228 → 204
german/*	210 → 197
guarani/numbers.ts	100 → 56

No defects in the source — one I introduced

Slicing guarani/numbers.ts's header left a duplicate */, which is a syntax error. The emit check caught it before push.

That's the third time this session the mechanical edit was itself the defect — after авторed in abkhaz and the half-replaced ★⚠ in cherokee/chuvash. Worth stating plainly: bulk comment rewriting has a defect rate of its own, and code-identity verification on every batch is what keeps it from shipping.

Where the argument is the value, not the fact

guarani/numbers.ts — cut 100 → 56, but the case survives intact. Without it, someone will "fix" the academic neologisms to the Spanish loans everyone actually speaks. Both halves kept:

• Estigarribia calls the system "purely of academic use. In the colloquial language, Spanish numerals are in common use beyond irundy '4'."
• and the Spanish route cannot be taken without inventing data — there is no attested Guaraní spelling of a Spanish numeral, because written Guaraní sidesteps the question with arabic digits.

Plus the fact that stops it reading as dead: 6–12 are lexicalised in the month names, so a speaker who says siete for 7 still says jasypokõi for July.

gujarati/normalize.ts — inheriting Hindi's symbol words is a silent deletion, not a wrong reading. They're written in Devanagari, which isn't in GUJARATI_WORD, so the tokenizer never emits them at all: "45%" → [pˈistalis]. That reframing is the whole reason the file exists, and it was buried under a defect inventory.

Also kept

• german — the ordinal detector fires on the following word or a licensing article, precisely so a sentence-final N. matches neither; and the two-form declension (am/im/des → weak -en, das/der/die → -e)
• greek — the \b warning, and the Latin↔Greek homoglyph class where a Latin o typed for the article ο makes the word vanish into the foreign path

Notation

3 inventory-boilerplate copies, 5 ★ → ⚠, gan's 🔷 spelled out.

Checks

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

🤖 Generated with Claude Code

https://claude.ai/code/session_012Lc3WnUgogC7okV7n53vjr