lyric-romanizer 0.2.0 → 0.3.0

package/CHANGELOG.md

@@ -0,0 +1,40 @@
+# Changelog
+
+## 0.3.0 — 2026-07-25
+
+Architecture pass: every piece of per-script knowledge now lives in one table per side of the light/heavy packaging seam, engines are lazy and injectable, and the universal fallback is observable. Public interface changes are **additive**; the two deliberate output-behavior changes are listed below with their rationale.
+
+### Added
+
+- **Engine adapters** — `createRomanizer({ engines })` overrides any built-in engine or plugs an adapter for externally-romanized scripts (`arabic`, `hebrew`, `malayalam`, `bengali`, `other`). The library still performs zero network I/O by default; `UnsupportedRomanizationError` is now precisely "no engine for this script".
+- **`RomanizeResult.fallbacks`** — per-line flags, `true` where the script engine failed and the line was universally transliterated as a last resort. Previously this degradation was silent and indistinguishable from success.
+- **`requiresExternalRomanization` on the detector subpath** — pure classification moved to the light, engine-free side; `lyric-romanizer/detector` can now answer the full pre-romanization triage (which script → is it latin → needs external) with zero engine payload. Still re-exported from the main entry.
+- **`RomanizeEngine`, `RomanizeEngineContext`, `ExternalScript`** exported types.
+- **Plain Node ESM support** — `import 'lyric-romanizer'` previously crashed outside bundlers (`@romanize/korean` named-export interop); engines now load through interop-safe dynamic imports.
+- **Lazy engine loading** — every engine loads on first use. Importing the main entry no longer executes any engine package; a Korean-only consumer never loads pinyin-pro or kuroshiro. Failed loads retry on the next call and degrade to the universal fallback (flagged) rather than crashing playback.
+
+### Changed — deliberate output-behavior changes
+
+1. **Latin guard under a pinned script.** `romanizeLines` pins one script for the whole array; lines with ASCII letters and no character of any detectable script are now returned unchanged instead of being fed to the pinned engine. Before: `romanizeLines(['Hello', '世界'])` → `['H e l l o', 'shì jiè']` (pinyin-pro spaces out Latin text). After: `['Hello', 'shì jiè']`. Rationale: the engine cannot romanize what isn't its script; the old output was mangled.
+2. **Out-of-contract script strings throw.** Untyped callers pinning a string outside `ScriptType` (e.g. `{ script: 'greek' }`) now get `UnsupportedRomanizationError` instead of a silent universal transliteration. This includes names inherited from `Object.prototype`: `{ script: 'toString' }` previously resolved to the inherited method and returned `'[object Undefined]'`. Unaffected: auto-detected scripts (detection only ever returns `ScriptType` members), and empty or letterless lines, which short-circuit to a no-op before any script is consulted — as they always did.
+
+All romanized output strings are byte-identical to 0.2.0 apart from those two cases, verified against a golden corpus across every supported script, dialect, preset, and error path. Two further notes for anyone diffing behavior:
+
+- `romanizeLine` with an engineless pinned script (e.g. `{ script: 'arabic' }`) still throws even when the line is pure Latin — the engine check precedes the latin guard, matching `romanizeLines`, which rejects such a script before looking at any line.
+- `NON_LATIN_SCRIPT_RE` is now derived from the range table, so its `.source` text changed: ranges are emitted in ascending code-point order, and the adjacent Hebrew and Arabic ranges collapse into one (8 groups → 7). The matching contract is unchanged — `.flags` is still `''` and `.test()` agrees with the old literal for every code point in the BMP, asserted by a test sweep. Only a consumer that snapshots, hashes, or string-compares the pattern text is affected; one that uses it as a matcher is not.
+
+### Fixed
+
+- README Quick Start example was impossible: `romanizeLines(['你好世界', 'こんにちは'])` returns `{ script: 'japanese' }` (kana is definitive), not `{ script: 'chinese' }` as documented. The example and the detection-granularity contract are now documented accurately.
+- Documented romanization outputs corrected against the actual engines (in all nine READMEs): Thai `สวัสดี` → `swasdi` (was `sawatdi`), Gurmukhi `ਨਮਸਤੇ` → `namasate` (was `namaste`; standard Punjabi spelling has no virama, so both consonants keep the inherent vowel), and the Kannada sample is now the correctly-spelled `ನಮಸ್ತೇ` → `namaste`. The feature list said "7 Indic scripts"; there are 6.
+- Dead defensive throws in the Sanscript branch (unreachable behind the catch-all fallback) removed; the silent `default:` switch arm is gone — the engine table is total over the script union, so a forgotten engine row is a **compile error** instead of silent garbage.
+- `NON_LATIN_SCRIPT_RE` is now derived from the same per-script range table that drives detection (previously a hand-maintained duplicate with a "keep aligned" comment). Verified byte-for-byte identical across the BMP.
+
+### Internal
+
+- Per-script knowledge concentrated: detection ranges + external classification in `SCRIPT_METADATA` (detector side), engine bindings in one total record (romanizer side). Adding a script is one row per side, compiler-enforced.
+- `selectCyrillicPreset` extracted as a pure, unit-tested function (the Ukrainian-detection headline feature previously had zero coverage).
+- Dependencies updated to latest (TypeScript 7, vitest 4.1.10, pinyin-pro 3.28.2). CI, Dependabot (daily, grouped), and a weekly latest-deps canary keep every engine on its newest release.
+- Test suite grew from 19 to 44 tests: injection seam, fallback observability, lazy-init retry, preset selection, latin guard, regex-derivation equivalence, tie-break regression, BMP-range invariant.
+- `pnpm run smoke` (`scripts/smoke.mjs`) exercises the built package under plain Node ESM: every engine's interop chain, the fallback path, both entry points, and an assertion that `dist/detector.js` imports nothing. Runs in CI and in the canary; `pnpm run verify` chains typecheck + build + test + smoke.
+- `package.json` declares `"sideEffects": false` (safe now that engines are lazy) and ships `src` so the published source maps resolve.

package/README.md

@@ -3,10 +3,28 @@
 [![npm version](https://img.shields.io/npm/v/lyric-romanizer.svg)](https://www.npmjs.com/package/lyric-romanizer)
 [![license](https://img.shields.io/npm/l/lyric-romanizer.svg)](https://github.com/thedavidweng/lyric-romanizer/blob/main/LICENSE)
 
+[English](https://github.com/thedavidweng/lyric-romanizer#readme) | [日本語](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.ja.md) | [中文（简体）](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.zh-CN.md) | [中文（粵語）](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.zh-yue.md) | [한국어](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.ko.md) | [Русский](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.ru.md) | [हिन्दी](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.hi.md) | [தமிழ்](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.ta.md) | [ไทย](https://github.com/thedavidweng/lyric-romanizer/blob/main/docs/README.th.md)
+
+> **Philosophy: Don't reinvent the wheel.**
+> This project deliberately avoids building romanization logic from scratch. Instead, it composes best-in-class, community-maintained libraries — one for each script — and focuses on the orchestration layer: script detection, engine routing, dialect handling, and a unified API. Every romanization engine in the dependency list is a dedicated, battle-tested library maintained by domain experts. That's the point.
+
 Script detection and local romanization engine for lyrics. Supports 12 scripts across Japanese, Chinese (Mandarin and Cantonese), Korean, Cyrillic, Indic, Tamil, and Thai — all running locally with zero API calls.
 
 Extracted from [Spotify Karaoke](https://github.com/haroldalan/spotify-karaoke). Used by [OpenKara](https://github.com/thedavidweng/openkara).
 
+## Features
+
+- **Zero API calls** — all romanization runs locally
+- **Auto script detection** — pass in text, get back the detected script
+- **12+ scripts** — Japanese, Chinese, Korean, Cyrillic, 6 Indic scripts, Tamil, Thai
+- **Cantonese support** — Jyutping alongside default Mandarin Pinyin
+- **Lightweight detector subpath** — import only script detection (and external-script classification) without pulling in romanization engines
+- **Lazy engines** — every engine loads on first use; importing the main entry costs nothing until you romanize
+- **Pluggable engines** — override any built-in engine, or plug your own adapter for externally-romanized scripts
+- **Observable fallbacks** — per-line flags tell you when an engine failed and a line was transliterated as a last resort
+- **Ukrainian-aware Cyrillic** — auto-detects Ukrainian-specific characters and applies the correct transliteration preset
+- **Runs in plain Node ESM** — no bundler required
+
 ## Installation
 
 ```bash
@@ -29,14 +47,16 @@ import { createRomanizer, detectScript } from 'lyric-romanizer';
 const romanizer = createRomanizer();
 
 // Auto-detect script and romanize
-const result = await romanizer.romanizeLines(['你好世界', 'こんにちは']);
-// { script: 'chinese', lines: ['nǐ hǎo shì jiè', 'こんにちは'] }
+const result = await romanizer.romanizeLines(['你好世界', '很高兴认识你']);
+// { script: 'chinese', lines: ['nǐ hǎo shì jiè', 'hěn gāo xìng rèn shi nǐ'], fallbacks: [false, false] }
 
 // Romanize a single line
 const line = await romanizer.romanizeLine('안녕하세요');
 // 'annyeonghaseyo'
 ```
 
+> **Detection granularity** — `romanizeLines` detects the dominant script **once across all lines** and pins it for every line (deliberate: kanji-only lines inside a Japanese song must reach the Japanese engine, and any kana in the array pins `japanese`). Calling `romanizeLine` in a loop instead detects **per line** and may route each line to a different engine. Pure-Latin lines under a pinned script are returned unchanged. See [Mixed-script arrays](#mixed-script-arrays).
+
 ## API
 
 ### Imports
@@ -52,7 +72,12 @@ import {
 } from 'lyric-romanizer';
 
 // Detector-only subpath — lightweight, no romanization dependencies
-import { detectScript, isLatinScript, NON_LATIN_SCRIPT_RE } from 'lyric-romanizer/detector';
+import {
+  detectScript,
+  isLatinScript,
+  requiresExternalRomanization,
+  NON_LATIN_SCRIPT_RE,
+} from 'lyric-romanizer/detector';
 ```
 
 ### Types
@@ -63,23 +88,33 @@ type ScriptType =
   | 'devanagari' | 'gujarati' | 'gurmukhi' | 'telugu'
   | 'kannada' | 'odia' | 'tamil' | 'malayalam'
   | 'bengali' | 'arabic' | 'hebrew' | 'thai'
-  | 'other';
+  | 'latin' | 'other';
 
 interface Romanizer {
   romanizeLine(line: string, options?: RomanizeOptions): Promise<string>;
   romanizeLines(lines: readonly string[], options?: RomanizeOptions): Promise<RomanizeResult>;
 }
 
+// `dialect` is only honored for 'chinese'; every other script ignores it.
 type RomanizeOptions = { script?: ScriptType; dialect?: 'mandarin' | 'cantonese' };
-type RomanizeResult = { script: ScriptType; lines: string[] };
-type RomanizerOptions = { japaneseDictPath?: string };
-```
 
-### Functions
+// `fallbacks` is aligned with `lines`: true where the engine failed and the
+// line was universally transliterated as a last resort.
+type RomanizeResult = { script: ScriptType; lines: string[]; fallbacks?: boolean[] };
 
-#### `createRomanizer(options?)`
+// An engine adapter: romanizes one line of its script. Throwing (or
+// rejecting) triggers the universal transliteration fallback.
+type RomanizeEngine = (line: string, context: { dialect: 'mandarin' | 'cantonese' }) => string | Promise<string>;
 
-Factory that returns a `Romanizer` instance. The Kuroshiro engine (Japanese) is lazily initialized on first use and cached.
+type RomanizerOptions = {
+  japaneseDictPath?: string;
+  engines?: Partial<Record<ScriptType, RomanizeEngine>>;
+};
+```
+
+### `createRomanizer(options?)`
+
+Factory that returns a `Romanizer` instance. Every engine is lazily loaded on first use and cached — a failed load retries on the next call.
 
 ```ts
 const romanizer = createRomanizer();
@@ -90,7 +125,27 @@ const romanizer = createRomanizer({
 });
 ```
 
-#### `detectScript(lines)`
+#### Engine adapters
+
+`options.engines` overrides the built-in engine for a script — or plugs an engine into a script that has none built in (`arabic`, `hebrew`, `malayalam`, `bengali`, `other`), so `romanizeLines` can handle every script through one interface. By default the library performs **zero network I/O**; plugging a remote adapter is an explicit caller decision.
+
+```ts
+const romanizer = createRomanizer({
+  engines: {
+    // Bring your own external romanization for scripts without a local engine.
+    arabic: async (line) => myTransliterationApi(line),
+    // Or replace a built-in engine (e.g. with a fake in tests).
+    korean: (line) => myKoreanRomanizer(line),
+  },
+});
+
+await romanizer.romanizeLines(['مرحبا']);
+// { script: 'arabic', lines: [...], fallbacks: [false] } — no longer throws
+```
+
+Scripts without an engine — built-in or injected — still throw `UnsupportedRomanizationError`.
+
+### `detectScript(lines)`
 
 Detects the dominant script in the given text lines. Checks for Japanese kana first (definitive), then scores all other scripts by character count.
 
@@ -102,7 +157,7 @@ detectScript(['Hello world']);        // 'latin'
 detectScript(['123 ???']);            // 'other'
 ```
 
-#### `isLatinScript(lines)`
+### `isLatinScript(lines)`
 
 Fast check — returns `true` if the text contains only Latin letters (no CJK, Cyrillic, Indic, etc.). Useful for skipping romanization entirely.
 
@@ -112,9 +167,9 @@ isLatinScript(['안녕하세요']);    // false
 isLatinScript(['♪♪♪']);         // false (no letters)
 ```
 
-#### `requiresExternalRomanization(script)`
+### `requiresExternalRomanization(script)`
 
-Returns `true` for scripts that cannot be romanized locally and require an external API.
+Returns `true` for scripts that have no built-in engine and require an external API. Importable from the lightweight `lyric-romanizer/detector` subpath, so answering "should I branch to an external service?" costs zero engine payload.
 
 ```ts
 requiresExternalRomanization('chinese');   // false
@@ -122,15 +177,13 @@ requiresExternalRomanization('arabic');    // true
 requiresExternalRomanization('malayalam'); // true
 ```
 
-### `Romanizer` Interface
-
-#### `romanizer.romanizeLine(line, options?)`
+### `romanizer.romanizeLine(line, options?)`
 
-Romanizes a single line. If `script` is omitted, it is auto-detected via `detectScript`. Returns the original line unchanged for Latin text or non-letter content.
+Romanizes a single line. If `script` is omitted, it is auto-detected **per line** via `detectScript` — looping over `romanizeLine` may route each line to a different engine, unlike `romanizeLines`, which pins one script for the whole array. Returns the original line unchanged for Latin text or non-letter content.
 
-For Chinese text, the `dialect` option controls the romanization system: `'mandarin'` (default) uses Pinyin, `'cantonese'` uses [Jyutping](https://github.com/CanCLID/to-jyutping).
+For Chinese text, the `dialect` option controls the romanization system: `'mandarin'` (default) uses Pinyin, `'cantonese'` uses [Jyutping](https://github.com/CanCLID/to-jyutping). Other scripts ignore `dialect`.
 
-**Throws** `UnsupportedRomanizationError` for external scripts.
+**Throws** `UnsupportedRomanizationError` for scripts without an engine (built-in or injected).
 
 ```ts
 await romanizer.romanizeLine('你好世界');
@@ -149,21 +202,21 @@ await romanizer.romanizeLine('مرحبا');
 // throws UnsupportedRomanizationError { script: 'arabic' }
 ```
 
-#### `romanizer.romanizeLines(lines, options?)`
+### `romanizer.romanizeLines(lines, options?)`
 
-Romanizes multiple lines in parallel. Returns the detected script and romanized lines.
+Romanizes multiple lines in parallel. Detects the dominant script **once across all lines** and pins it for every line (see [Mixed-script arrays](#mixed-script-arrays)). Returns the script, the romanized lines, and per-line `fallbacks` flags — `true` where the engine failed and the line was universally transliterated as a last resort.
 
 ```ts
-const { script, lines } = await romanizer.romanizeLines([
+const { script, lines, fallbacks } = await romanizer.romanizeLines([
   'สวัสดี',
   'ชาวโลก',
 ]);
-// { script: 'thai', lines: ['sawatdi', 'chaolok'] }
+// { script: 'thai', lines: ['swasdi', 'chaolok'], fallbacks: [false, false] }
 ```
 
 ### `UnsupportedRomanizationError`
 
-Thrown when attempting to romanize a script that requires an external API. Has a `script` property for programmatic handling.
+Thrown when attempting to romanize a script that has no engine — built in or injected via `options.engines`. Has a `script` property for programmatic handling.
 
 ```ts
 try {
@@ -190,12 +243,12 @@ try {
 | Cyrillic | [cyrillic-to-translit-js](https://github.com/greybax/CyrillicToTranslitJS) | `Привет` → `Privet` |
 | Devanagari | [sanscript](https://github.com/indic-transliteration/sanscript) | `नमस्ते` → `namaste` |
 | Gujarati | [sanscript](https://github.com/indic-transliteration/sanscript) | `નમસ્તે` → `namaste` |
-| Gurmukhi | [sanscript](https://github.com/indic-transliteration/sanscript) | `ਨਮਸਤੇ` → `namaste` |
+| Gurmukhi | [sanscript](https://github.com/indic-transliteration/sanscript) | `ਨਮਸਤੇ` → `namasate` |
 | Telugu | [sanscript](https://github.com/indic-transliteration/sanscript) | `నమస్తే` → `namaste` |
-| Kannada | [sanscript](https://github.com/indic-transliteration/sanscript) | `నಮస్తె` → `namaste` |
+| Kannada | [sanscript](https://github.com/indic-transliteration/sanscript) | `ನಮಸ್ತೇ` → `namaste` |
 | Odia | [sanscript](https://github.com/indic-transliteration/sanscript) | `ନମସ୍ତେ` → `namaste` |
 | Tamil | [tamil-romanizer](https://github.com/haroldalan/tamil-romanizer) | `வணக்கம்` → `vanakkam` |
-| Thai | [@dehoist/romanize-thai](https://github.com/Dehoist/Open-Source) | `สวัสดี` → `sawatdi` |
+| Thai | [@dehoist/romanize-thai](https://github.com/Dehoist/Open-Source) | `สวัสดี` → `swasdi` |
 
 ### External (requires API)
 
@@ -207,13 +260,26 @@ try {
 | Hebrew | Google Translate `dt=rm` |
 | Other | Google Translate `dt=rm` |
 
-Use `requiresExternalRomanization()` to detect these and branch to your preferred API.
+Use `requiresExternalRomanization()` to detect these and branch to your preferred API — or plug the API in once as an [engine adapter](#engine-adapters) and let `romanizeLines` handle every script.
+
+## Script-Specific Notes
+
+### Mixed-script arrays
+
+`romanizeLines` pins the **dominant** script of the whole array to every line. This is deliberate: a kanji-only line inside a Japanese song is indistinguishable from Chinese on its own (kanji and hanzi share the same Unicode block), so only whole-array context routes it to the right engine. Consequences worth knowing:
+
+- Any kana anywhere in the array pins the whole array to `japanese` — kana is definitive proof.
+- A line of a *different* non-Latin script inside the array is still fed to the pinned engine.
+- Pure-Latin lines (an English chorus inside a CJK song) are **returned unchanged** instead of being fed to the pinned engine.
+- Per-line `fallbacks` flags report when an engine failed and a line degraded to universal transliteration.
+
+If you need true per-line engine routing, call `romanizeLine` in a loop — it detects per line.
 
-## Cyrillic Detection
+### Cyrillic Detection
 
 Cyrillic auto-detects Ukrainian-specific characters (`і`, `ї`, `є`, `ґ`) and applies the Ukrainian transliteration preset. All other Cyrillic text defaults to Russian.
 
-## Cantonese Support
+### Cantonese Support
 
 Chinese text defaults to Mandarin (Pinyin). Pass `dialect: 'cantonese'` in `RomanizeOptions` to romanize Chinese text to [Jyutping](https://github.com/CanCLID/to-jyutping) instead.
 

package/dist/detector.d.ts

@@ -1,5 +1,79 @@
 import type { ScriptType } from './types.js';
+declare const SCRIPT_METADATA: {
+    readonly japanese: {
+        readonly ranges: readonly [readonly [12352, 12543]];
+        readonly definitive: true;
+    };
+    readonly chinese: {
+        readonly ranges: readonly [readonly [19968, 40959]];
+    };
+    readonly korean: {
+        readonly ranges: readonly [readonly [44032, 55215]];
+    };
+    readonly cyrillic: {
+        readonly ranges: readonly [readonly [1024, 1279]];
+    };
+    readonly devanagari: {
+        readonly ranges: readonly [readonly [2304, 2431]];
+    };
+    readonly gujarati: {
+        readonly ranges: readonly [readonly [2688, 2815]];
+    };
+    readonly gurmukhi: {
+        readonly ranges: readonly [readonly [2560, 2687]];
+    };
+    readonly telugu: {
+        readonly ranges: readonly [readonly [3072, 3199]];
+    };
+    readonly kannada: {
+        readonly ranges: readonly [readonly [3200, 3327]];
+    };
+    readonly odia: {
+        readonly ranges: readonly [readonly [2816, 2943]];
+    };
+    readonly tamil: {
+        readonly ranges: readonly [readonly [2944, 3071]];
+    };
+    readonly malayalam: {
+        readonly ranges: readonly [readonly [3328, 3455]];
+        readonly external: true;
+    };
+    readonly bengali: {
+        readonly ranges: readonly [readonly [2432, 2559]];
+        readonly external: true;
+    };
+    readonly arabic: {
+        readonly ranges: readonly [readonly [1536, 1791]];
+        readonly external: true;
+    };
+    readonly hebrew: {
+        readonly ranges: readonly [readonly [1424, 1535]];
+        readonly external: true;
+    };
+    readonly thai: {
+        readonly ranges: readonly [readonly [3584, 3711]];
+    };
+    readonly latin: {
+        readonly ranges: readonly [];
+    };
+    readonly other: {
+        readonly ranges: readonly [];
+        readonly external: true;
+    };
+};
+/**
+ * Scripts classified as external in SCRIPT_METADATA. The engine table in
+ * romanizer.ts excludes exactly these members, so reclassifying a script here
+ * surfaces as a compile error there.
+ */
+export type ExternalScript = {
+    [K in ScriptType]: (typeof SCRIPT_METADATA)[K] extends {
+        external: true;
+    } ? K : never;
+}[ScriptType];
 export declare const NON_LATIN_SCRIPT_RE: RegExp;
 export declare function isLatinScript(lines: readonly string[]): boolean;
 export declare function detectScript(lines: readonly string[]): ScriptType;
+export declare function requiresExternalRomanization(script: ScriptType): boolean;
+export {};
 //# sourceMappingURL=detector.d.ts.map

package/dist/detector.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"detector.d.ts","sourceRoot":"","sources":["../src/detector.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAwB7C,eAAO,MAAM,mBAAmB,QAC8E,CAAC;AAE/G,wBAAgB,aAAa,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,GAAG,OAAO,CAG/D;AAED,wBAAgB,YAAY,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,GAAG,UAAU,CAcjE"}
+{"version":3,"file":"detector.d.ts","sourceRoot":"","sources":["../src/detector.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAwB7C,QAAA,MAAM,eAAe;;iBAGP,MAAM;iBAAsB,UAAU;;;iBACvC,MAAM;;;iBACP,MAAM;;;iBACJ,MAAM;;;iBACJ,MAAM;;;iBACR,MAAM;;;iBACN,MAAM;;;iBACR,MAAM;;;iBACL,MAAM;;;iBACT,MAAM;;;iBACL,MAAM;;;iBACF,MAAM;iBAAsB,QAAQ;;;iBACtC,MAAM;iBAAsB,QAAQ;;;iBACrC,MAAM;iBAAsB,QAAQ;;;iBACpC,MAAM;iBAAsB,QAAQ;;;iBACtC,MAAM;;;iBACL,MAAM;;;iBACN,MAAM;iBAAM,QAAQ;;CACoB,CAAC;AAEpD;;;;GAIG;AACH,MAAM,MAAM,cAAc,GAAG;KAC1B,CAAC,IAAI,UAAU,GAAG,CAAC,OAAO,eAAe,CAAC,CAAC,CAAC,CAAC,SAAS;QAAE,QAAQ,EAAE,IAAI,CAAA;KAAE,GAAG,CAAC,GAAG,KAAK;CACtF,CAAC,UAAU,CAAC,CAAC;AAwCd,eAAO,MAAM,mBAAmB,QAE/B,CAAC;AAEF,wBAAgB,aAAa,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,GAAG,OAAO,CAG/D;AAED,wBAAgB,YAAY,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,GAAG,UAAU,CAejE;AAED,wBAAgB,4BAA4B,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAMxE"}

package/dist/detector.js

@@ -1,33 +1,75 @@
 const LETTER_RE = /\p{L}/u;
-const SCRIPT_SCORES = [
-    ['chinese', /[\u4E00-\u9FFF]/g],
-    ['korean', /[\uAC00-\uD7AF]/g],
-    ['cyrillic', /[\u0400-\u04FF]/g],
-    ['devanagari', /[\u0900-\u097F]/g],
-    ['gujarati', /[\u0A80-\u0AFF]/g],
-    ['gurmukhi', /[\u0A00-\u0A7F]/g],
-    ['telugu', /[\u0C00-\u0C7F]/g],
-    ['kannada', /[\u0C80-\u0CFF]/g],
-    ['odia', /[\u0B00-\u0B7F]/g],
-    ['tamil', /[\u0B80-\u0BFF]/g],
-    ['malayalam', /[\u0D00-\u0D7F]/g],
-    ['bengali', /[\u0980-\u09FF]/g],
-    ['arabic', /[\u0600-\u06FF]/g],
-    ['hebrew', /[\u0590-\u05FF]/g],
-    ['thai', /[\u0E00-\u0E7F]/g],
-];
-// Keep this range aligned with extension content script behavior for the
-// "latin fast path" check used during mode switching.
-export const NON_LATIN_SCRIPT_RE = /[\u3040-\u30FF\u4E00-\u9FFF\uAC00-\uD7AF\u0400-\u04FF\u0900-\u0D7F\u0600-\u06FF\u0590-\u05FF\u0E00-\u0E7F]/;
+// Single source of truth for per-script knowledge on the light (engine-free)
+// side of the packaging seam: detection ranges and local/external
+// classification. SCRIPT_SCORES, NON_LATIN_SCRIPT_RE, and
+// requiresExternalRomanization are all derived from this table, and the
+// engine table in romanizer.ts is compile-time checked against the
+// classification, so adding a script here is the single detector-side edit.
+//
+// Entry order is load-bearing: it is the tie-break priority of detectScript —
+// on an equal character count, the earlier entry wins.
+const SCRIPT_METADATA = {
+    // Kana is definitive: kanji shares the Han block with chinese, so any kana
+    // must decide 'japanese' before Han scoring runs.
+    japanese: { ranges: [[0x3040, 0x30ff]], definitive: true },
+    chinese: { ranges: [[0x4e00, 0x9fff]] },
+    korean: { ranges: [[0xac00, 0xd7af]] },
+    cyrillic: { ranges: [[0x0400, 0x04ff]] },
+    devanagari: { ranges: [[0x0900, 0x097f]] },
+    gujarati: { ranges: [[0x0a80, 0x0aff]] },
+    gurmukhi: { ranges: [[0x0a00, 0x0a7f]] },
+    telugu: { ranges: [[0x0c00, 0x0c7f]] },
+    kannada: { ranges: [[0x0c80, 0x0cff]] },
+    odia: { ranges: [[0x0b00, 0x0b7f]] },
+    tamil: { ranges: [[0x0b80, 0x0bff]] },
+    malayalam: { ranges: [[0x0d00, 0x0d7f]], external: true },
+    bengali: { ranges: [[0x0980, 0x09ff]], external: true },
+    arabic: { ranges: [[0x0600, 0x06ff]], external: true },
+    hebrew: { ranges: [[0x0590, 0x05ff]], external: true },
+    thai: { ranges: [[0x0e00, 0x0e7f]] },
+    latin: { ranges: [] },
+    other: { ranges: [], external: true },
+};
+function escapeCodePoint(codePoint) {
+    return `\\u${codePoint.toString(16).toUpperCase().padStart(4, '0')}`;
+}
+function characterClass(ranges) {
+    return ranges
+        .map(([start, end]) => (start === end ? escapeCodePoint(start) : `${escapeCodePoint(start)}-${escapeCodePoint(end)}`))
+        .join('');
+}
+function mergeRanges(ranges) {
+    const sorted = [...ranges].sort((a, b) => a[0] - b[0]);
+    const merged = [];
+    for (const [start, end] of sorted) {
+        const last = merged[merged.length - 1];
+        if (last && start <= last[1] + 1) {
+            last[1] = Math.max(last[1], end);
+        }
+        else {
+            merged.push([start, end]);
+        }
+    }
+    return merged;
+}
+const METADATA_ENTRIES = Object.entries(SCRIPT_METADATA);
+const DEFINITIVE_SCRIPTS = METADATA_ENTRIES.filter(([, meta]) => meta.definitive).map(([script, meta]) => [script, new RegExp(`[${characterClass(meta.ranges)}]`)]);
+const SCRIPT_SCORES = METADATA_ENTRIES.filter(([, meta]) => meta.ranges.length > 0 && !meta.definitive).map(([script, meta]) => [script, new RegExp(`[${characterClass(meta.ranges)}]`, 'g')]);
+// Derived union of every detectable range. Also consumed by the extension
+// content script's "latin fast path" during mode switching — deriving it from
+// SCRIPT_METADATA keeps it aligned with detection structurally instead of by
+// comment.
+export const NON_LATIN_SCRIPT_RE = new RegExp(`[${characterClass(mergeRanges(METADATA_ENTRIES.flatMap(([, meta]) => meta.ranges)))}]`);
 export function isLatinScript(lines) {
     const text = lines.join('');
     return !NON_LATIN_SCRIPT_RE.test(text) && LETTER_RE.test(text);
 }
 export function detectScript(lines) {
     const text = lines.join('');
-    // Japanese: any kana is definitive and must be checked before CJK scoring.
-    if (/[\u3040-\u30FF]/.test(text))
-        return 'japanese';
+    for (const [script, pattern] of DEFINITIVE_SCRIPTS) {
+        if (pattern.test(text))
+            return script;
+    }
     let best = ['other', 0];
     for (const [script, pattern] of SCRIPT_SCORES) {
         const score = (text.match(pattern) ?? []).length;
@@ -38,4 +80,12 @@ export function detectScript(lines) {
         return best[0];
     return LETTER_RE.test(text) ? 'latin' : 'other';
 }
+export function requiresExternalRomanization(script) {
+    // Own-property check so an out-of-contract string from untyped callers
+    // (including inherited names like 'toString') returns false rather than
+    // resolving up the prototype chain.
+    if (!Object.hasOwn(SCRIPT_METADATA, script))
+        return false;
+    return SCRIPT_METADATA[script].external === true;
+}
 //# sourceMappingURL=detector.js.map

package/dist/detector.js.map

@@ -1 +1 @@
-{"version":3,"file":"detector.js","sourceRoot":"","sources":["../src/detector.ts"],"names":[],"mappings":"AAEA,MAAM,SAAS,GAAG,QAAQ,CAAC;AAE3B,MAAM,aAAa,GAAyC;IAC1D,CAAC,SAAS,EAAE,kBAAkB,CAAC;IAC/B,CAAC,QAAQ,EAAE,kBAAkB,CAAC;IAC9B,CAAC,UAAU,EAAE,kBAAkB,CAAC;IAChC,CAAC,YAAY,EAAE,kBAAkB,CAAC;IAClC,CAAC,UAAU,EAAE,kBAAkB,CAAC;IAChC,CAAC,UAAU,EAAE,kBAAkB,CAAC;IAChC,CAAC,QAAQ,EAAE,kBAAkB,CAAC;IAC9B,CAAC,SAAS,EAAE,kBAAkB,CAAC;IAC/B,CAAC,MAAM,EAAE,kBAAkB,CAAC;IAC5B,CAAC,OAAO,EAAE,kBAAkB,CAAC;IAC7B,CAAC,WAAW,EAAE,kBAAkB,CAAC;IACjC,CAAC,SAAS,EAAE,kBAAkB,CAAC;IAC/B,CAAC,QAAQ,EAAE,kBAAkB,CAAC;IAC9B,CAAC,QAAQ,EAAE,kBAAkB,CAAC;IAC9B,CAAC,MAAM,EAAE,kBAAkB,CAAC;CAC7B,CAAC;AAEF,yEAAyE;AACzE,sDAAsD;AACtD,MAAM,CAAC,MAAM,mBAAmB,GAC9B,4GAA4G,CAAC;AAE/G,MAAM,UAAU,aAAa,CAAC,KAAwB;IACpD,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC5B,OAAO,CAAC,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAwB;IACnD,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAE5B,2EAA2E;IAC3E,IAAI,iBAAiB,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,UAAU,CAAC;IAEpD,IAAI,IAAI,GAAkC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;IACvD,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,aAAa,EAAE,CAAC;QAC9C,MAAM,KAAK,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC;QACjD,IAAI,KAAK,GAAG,IAAI,CAAC,CAAC,CAAC;YAAE,IAAI,GAAG,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IAC9C,CAAC;IAED,IAAI,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC;QAAE,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC;IAChC,OAAO,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC;AAClD,CAAC"}
+{"version":3,"file":"detector.js","sourceRoot":"","sources":["../src/detector.ts"],"names":[],"mappings":"AAEA,MAAM,SAAS,GAAG,QAAQ,CAAC;AAa3B,6EAA6E;AAC7E,kEAAkE;AAClE,0DAA0D;AAC1D,wEAAwE;AACxE,mEAAmE;AACnE,4EAA4E;AAC5E,EAAE;AACF,8EAA8E;AAC9E,uDAAuD;AACvD,MAAM,eAAe,GAAG;IACtB,2EAA2E;IAC3E,kDAAkD;IAClD,QAAQ,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE,UAAU,EAAE,IAAI,EAAE;IAC1D,OAAO,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACvC,MAAM,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACtC,QAAQ,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACxC,UAAU,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IAC1C,QAAQ,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACxC,QAAQ,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACxC,MAAM,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACtC,OAAO,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACvC,IAAI,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACpC,KAAK,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACrC,SAAS,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE;IACzD,OAAO,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE;IACvD,MAAM,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE;IACtD,MAAM,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE;IACtD,IAAI,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,EAAE;IACpC,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;IACrB,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE;CACY,CAAC;AAWpD,SAAS,eAAe,CAAC,SAAiB;IACxC,OAAO,MAAM,SAAS,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;AACvE,CAAC;AAED,SAAS,cAAc,CAAC,MAA4B;IAClD,OAAO,MAAM;SACV,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,GAAG,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,KAAK,GAAG,CAAC,CAAC,CAAC,eAAe,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,eAAe,CAAC,KAAK,CAAC,IAAI,eAAe,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;SACrH,IAAI,CAAC,EAAE,CAAC,CAAC;AACd,CAAC;AAED,SAAS,WAAW,CAAC,MAA4B;IAC/C,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACvD,MAAM,MAAM,GAA4B,EAAE,CAAC;IAC3C,KAAK,MAAM,CAAC,KAAK,EAAE,GAAG,CAAC,IAAI,MAAM,EAAE,CAAC;QAClC,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACvC,IAAI,IAAI,IAAI,KAAK,IAAI,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;YACjC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;QACnC,CAAC;aAAM,CAAC;YACN,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,CAAC;QAC5B,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,gBAAgB,GAAG,MAAM,CAAC,OAAO,CAAC,eAAe,CAAoC,CAAC;AAE5F,MAAM,kBAAkB,GAAyC,gBAAgB,CAAC,MAAM,CACtF,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAC9B,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,IAAI,MAAM,CAAC,IAAI,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAU,CAAC,CAAC;AAE7F,MAAM,aAAa,GAAyC,gBAAgB,CAAC,MAAM,CACjF,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CACzD,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,IAAI,MAAM,CAAC,IAAI,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,CAAU,CAAC,CAAC;AAElG,0EAA0E;AAC1E,8EAA8E;AAC9E,6EAA6E;AAC7E,WAAW;AACX,MAAM,CAAC,MAAM,mBAAmB,GAAG,IAAI,MAAM,CAC3C,IAAI,cAAc,CAAC,WAAW,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,CACxF,CAAC;AAEF,MAAM,UAAU,aAAa,CAAC,KAAwB;IACpD,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC5B,OAAO,CAAC,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAwB;IACnD,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAE5B,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,kBAAkB,EAAE,CAAC;QACnD,IAAI,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;YAAE,OAAO,MAAM,CAAC;IACxC,CAAC;IAED,IAAI,IAAI,GAAkC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;IACvD,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,aAAa,EAAE,CAAC;QAC9C,MAAM,KAAK,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC;QACjD,IAAI,KAAK,GAAG,IAAI,CAAC,CAAC,CAAC;YAAE,IAAI,GAAG,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IAC9C,CAAC;IAED,IAAI,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC;QAAE,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC;IAChC,OAAO,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC;AAClD,CAAC;AAED,MAAM,UAAU,4BAA4B,CAAC,MAAkB;IAC7D,uEAAuE;IACvE,wEAAwE;IACxE,oCAAoC;IACpC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,EAAE,MAAM,CAAC;QAAE,OAAO,KAAK,CAAC;IAC1D,OAAQ,eAA8C,CAAC,MAAM,CAAC,CAAC,QAAQ,KAAK,IAAI,CAAC;AACnF,CAAC"}

package/dist/index.d.ts

@@ -1,4 +1,4 @@
-export { detectScript, isLatinScript, NON_LATIN_SCRIPT_RE } from './detector.js';
-export { createRomanizer, requiresExternalRomanization } from './romanizer.js';
-export { UnsupportedRomanizationError, type RomanizeOptions, type RomanizeResult, type Romanizer, type RomanizerOptions, type ScriptType, } from './types.js';
+export { detectScript, isLatinScript, requiresExternalRomanization, NON_LATIN_SCRIPT_RE, type ExternalScript, } from './detector.js';
+export { createRomanizer } from './romanizer.js';
+export { UnsupportedRomanizationError, type RomanizeEngine, type RomanizeEngineContext, type RomanizeOptions, type RomanizeResult, type Romanizer, type RomanizerOptions, type ScriptType, } from './types.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AACjF,OAAO,EAAE,eAAe,EAAE,4BAA4B,EAAE,MAAM,gBAAgB,CAAC;AAC/E,OAAO,EACL,4BAA4B,EAC5B,KAAK,eAAe,EACpB,KAAK,cAAc,EACnB,KAAK,SAAS,EACd,KAAK,gBAAgB,EACrB,KAAK,UAAU,GAChB,MAAM,YAAY,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,YAAY,EACZ,aAAa,EACb,4BAA4B,EAC5B,mBAAmB,EACnB,KAAK,cAAc,GACpB,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EACL,4BAA4B,EAC5B,KAAK,cAAc,EACnB,KAAK,qBAAqB,EAC1B,KAAK,eAAe,EACpB,KAAK,cAAc,EACnB,KAAK,SAAS,EACd,KAAK,gBAAgB,EACrB,KAAK,UAAU,GAChB,MAAM,YAAY,CAAC"}

package/dist/index.js

@@ -1,4 +1,4 @@
-export { detectScript, isLatinScript, NON_LATIN_SCRIPT_RE } from './detector.js';
-export { createRomanizer, requiresExternalRomanization } from './romanizer.js';
+export { detectScript, isLatinScript, requiresExternalRomanization, NON_LATIN_SCRIPT_RE, } from './detector.js';
+export { createRomanizer } from './romanizer.js';
 export { UnsupportedRomanizationError, } from './types.js';
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AACjF,OAAO,EAAE,eAAe,EAAE,4BAA4B,EAAE,MAAM,gBAAgB,CAAC;AAC/E,OAAO,EACL,4BAA4B,GAM7B,MAAM,YAAY,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,YAAY,EACZ,aAAa,EACb,4BAA4B,EAC5B,mBAAmB,GAEpB,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EACL,4BAA4B,GAQ7B,MAAM,YAAY,CAAC"}

package/dist/romanizer.d.ts

@@ -1,4 +1,9 @@
-import type { Romanizer, RomanizerOptions, ScriptType } from './types.js';
-export declare function requiresExternalRomanization(script: ScriptType): boolean;
+import type { Romanizer, RomanizerOptions } from './types.js';
+/**
+ * Ukrainian-specific characters pick the Ukrainian preset; all other Cyrillic
+ * romanizes as Russian. Applied per line, so a mixed-language song switches
+ * preset line by line.
+ */
+export declare function selectCyrillicPreset(line: string): 'ru' | 'uk';
 export declare function createRomanizer(options?: RomanizerOptions): Romanizer;
 //# sourceMappingURL=romanizer.d.ts.map

package/dist/romanizer.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"romanizer.d.ts","sourceRoot":"","sources":["../src/romanizer.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAGV,SAAS,EACT,gBAAgB,EAChB,UAAU,EACX,MAAM,YAAY,CAAC;AAwBpB,wBAAgB,4BAA4B,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAExE;AA+GD,wBAAgB,eAAe,CAAC,OAAO,CAAC,EAAE,gBAAgB,GAAG,SAAS,CAErE"}
+{"version":3,"file":"romanizer.d.ts","sourceRoot":"","sources":["../src/romanizer.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAKV,SAAS,EACT,gBAAgB,EAEjB,MAAM,YAAY,CAAC;AA4EpB;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,GAAG,IAAI,CAE9D;AA+ID,wBAAgB,eAAe,CAAC,OAAO,CAAC,EAAE,gBAAgB,GAAG,SAAS,CAErE"}