/**
 * The scripts `foldName` and `normalize` keep, alongside `[a-z0-9]`, instead
 * of collapsing to a space/dropping. **The one source both import from** — do
 * not inline this pattern a second time, for the same reason `fold-name.ts`
 * itself warns against a second private fold: a table difference between
 * copies turns a match into a conflict.
 *
 * ## Why an allowlist and not `\p{L}\p{N}`
 *
 * Three variants were measured against every stored Team/League/Venue/City
 * name (57,747 Team/League/Venue names, 274,101 City names):
 *
 * | variant | Latin folds that change |
 * |---|---|
 * | `\p{L}\p{N}` + NFD | 13 — `\p{L}` matches `º`, so `1º de Maio` folds to
 * `"1 de maio"` losing the `º` as a space rather than keeping it attached;
 * `\p{N}` matches `³` in `MAC³PARK` |
 * | `\p{L}\p{N}` + NFKD | 13 — better (`º` → `o`) but still a change |
 * | **this allowlist** | **0** |
 *
 * A general Unicode-letter class touches punctuation-like letters (ordinal
 * indicators, superscripts) that happen to be classified `\p{L}`/`\p{N}` but
 * are not part of a name the way a script's own letters are. An explicit
 * allowlist only ever ADDS scripts we deliberately chose, so it cannot
 * regress a fold that already works — which is what makes it safe to widen a
 * function with over a dozen consumers (`league-aliases`, `country-matcher`,
 * `opta-league-match`, `opta-team-match`, `venue-claim-verification`,
 * `team-duplicate-guard`, city matching, …).
 *
 * ## Why these ten scripts and no others
 *
 * They cover the 18 locales the UI ships (`weekendgoals-ui-next/src/i18n/
 * request.ts`) plus Thai and Devanagari, where GeoNames' alternate-name
 * coverage is dense enough to be worth keeping. **Do not add a script
 * speculatively** — every script added is index entries in every locale, on
 * every Team/League/Venue/City row.
 */
/**
 * `゙゚` — the Katakana-Hiragana voiced/semi-voiced sound marks
 * (dakuten/handakuten) — are listed by codepoint, not as a script.
 * `normalize("NFD")` decomposes a precomposed kana like バ (ba) into
 * `30CF 3099` (ハ + dakuten) before this pattern runs, and Unicode classifies
 * every combining mark as `\p{Script=Inherited}` regardless of which script
 * it visually modifies — NOT as `\p{Script=Katakana}`. Adding
 * `\p{Script=Inherited}` wholesale was tried and reverted: it also admits
 * the zero-width joiner/non-joiner (U+200C/200D) used inside Latin
 * transliterations of Pashto/Dari administrative names (e.g. GeoNames'
 * `ʿlaqh‌dary`), which glued two words together with an invisible character
 * instead of the space this allowlist used to leave — a real Latin
 * regression the exhaustive check below caught. Listing just the two
 * codepoints Japanese needs avoids that: without them `バルセロナ` silently
 * becomes `ハルセロナ` — a different word, not a lost accent, since バ→ハ is
 * a change of consonant, not of accent. Latin accents need no such carve-out:
 * they are already gone by the time this pattern runs, stripped by the
 * dedicated `̀-ͯ` pass both `fold-name.ts` and `normalize.ts`
 * run first. (Arabic's harakat are `Inherited` too and remain unhandled —
 * they were already stripped before this change, by both the old code and
 * the new, so leaving them out is unchanged behaviour, not a regression.)
 */
/**
 * **The safety measurement, re-run 2026-08-27 after the allowlist moved from
 * two Japanese codepoints to `\p{Mn}\p{Mc}` plus U+30FC.**
 *
 * Over 1,131,171 distinct names across Teams, Leagues, Venues and Cities —
 * 802,869 of them Latin-only — **2 Latin folds change**, and 301,371 non-Latin
 * names gain a fold they did not have.
 *
 * The two are `ga׳leg׳wleyah` and `nyz׳ny stwdny`: GeoNames romanisations that
 * use the Hebrew geresh (U+05F3) as a word separator, where it now survives
 * instead of becoming a space. The mark is kept deliberately — in Hebrew the
 * geresh is word-INTERNAL, forming letters Hebrew has no glyph for (ג׳ = "j")
 * — so this is a real trade rather than an oversight: two obscure Latin
 * transliterations against every Hebrew name that contains one. An earlier
 * pass claimed zero; do not restate that number without re-running the count.
 */
export declare const SCRIPT_ALLOWLIST: string;
//# sourceMappingURL=script-allowlist.d.ts.map