[rathesDot]

Words with Grantha letters came out half in Tamil script: getIso15919('ஜன்னல்') gave ஜṉṉal, and getIPA('சந்தோஷம்') gave t͡ɕan̪toːஷm. Two reasons:

• grantha.json held only the six bare consonants (ஜ், ஷ், …). A Grantha letter with a vowel (ஜ, ஜா, ஷோ) was never mapped, even with { grantha: true }.
• Grantha was off unless the caller asked for it. learn-tamil.com never did, so some live dictionary titles show Tamil letters, for example tuஷ்pirayōkam.

Change

• grantha.json has the same shape as script.json: one row per Grantha consonant (ஜ், ஶ், ஷ், ஸ், ஹ், க்ஷ்), with the dead consonant followed by its 12 vowel forms. index continues from script.json (247–324). I generated the rows from the six existing consonants and script.json's vowel row, nothing typed by hand.
• Grantha is on by default. { grantha: false } keeps the old behaviour. The only change in index.ts is that default; the lookup is unchanged.
• Tests: the granthaScript snapshot and its shape (6 rows × 13), plus words with Grantha syllables.
• README: updated to match.

Tamil	ISO 15919	IPA
ஜன்னல்	jaṉṉal	d͡ʑannal
சந்தோஷம்	cantōṣam	t͡ɕan̪toːʂam
கஷ்டம்	kaṣṭam	kaʂɖam
ஹோட்டல்	hōṭṭal	hoːɖɖal
பக்ஷி	pakṣi	pakʂi

Breaking

• Default output changes: text with Grantha letters is now transliterated without the option.
• Data shape changes: granthaScript rows grow from 1 entry to 13.

So this should be released as 1.0.0. I'll open the release pull request once this is merged, as CLAUDE.md describes. learn-tamil.com only looks letters up with .flat().find(...), so the shape change doesn't affect it.

Note: ஸ்ரீ gives srī, not the conventional śrī. It's written with ஸ (s), and I didn't add a special case.

Checks

pnpm test (8 passed, script snapshot unchanged, new granthaScript snapshot), pnpm eslint, pnpm tsc --noEmit and pnpm format --check all pass.