@stll/anonymize-wasm 1.4.8 → 1.4.10

package/dist/wasm.mjs

@@ -33,6 +33,10 @@ const isLegalFormsEnabled = (config) => config.enableLegalForms !== false;
 * shallow copies (spread). Uses position + label so the
 * key is identical for the original object and any
 * `{ ...entity }` copy produced by mergeAndDedup.
+*
+* @deprecated No longer used internally: coref alias
+*   links travel on the entities themselves
+*   (`corefSourceText`). Kept for API compatibility.
 */
 const corefKey = (e) => `${e.start}:${e.end}:${e.label}`;
 /** Create a fresh, empty pipeline context. */
@@ -62,8 +66,7 @@ const createPipelineContext = () => ({
 	roleStopSetPromise: null,
 	zoneHeadingPatterns: null,
 	zoneSigningPatterns: null,
-	zoneInitPromise: null,
-	corefSourceMap: /* @__PURE__ */ new Map()
+	zoneInitPromise: null
 });
 /**
 * Module-level default context. Used when callers
@@ -1761,11 +1764,14 @@ const isWordChar = (ch) => {
 * character before the start and after the end are NOT
 * word characters (letter/digit).
 *
-* Populates `ctx.corefSourceMap` with entries linking
-* each coref entity to its source entity text, for
-* consistent placeholder numbering.
+* Each returned alias carries `corefSourceText` linking
+* it to its source entity text, for consistent
+* placeholder numbering.
+*
+* @param _ctx Unused. Kept for signature compatibility;
+*   alias links now travel on the entities themselves.
 */
-const findCoreferenceSpans = (fullText, terms, ctx = defaultContext) => {
+const findCoreferenceSpans = (fullText, terms, _ctx = defaultContext) => {
 	const results = [];
 	for (const term of terms) {
 		let searchFrom = 0;
@@ -1779,16 +1785,15 @@ const findCoreferenceSpans = (fullText, terms, ctx = defaultContext) => {
 				searchFrom = idx + 1;
 				continue;
 			}
-			const entity = {
+			results.push({
 				start: idx,
 				end: matchEnd,
 				label: term.label,
 				text: term.alias,
 				score: .95,
-				source: DETECTION_SOURCES.COREFERENCE
-			};
-			ctx.corefSourceMap.set(corefKey(entity), term.sourceText);
-			results.push(entity);
+				source: DETECTION_SOURCES.COREFERENCE,
+				corefSourceText: term.sourceText
+			});
 			searchFrom = matchEnd;
 		}
 	}
@@ -7954,6 +7959,308 @@ const processCountryMatches = (allMatches, sliceStart, sliceEnd, fullText, data)
 	return results;
 };
 //#endregion
+//#region src/config/titles.ts
+/**
+* Academic and professional title prefixes.
+* Plain text; the detector auto-escapes for regex.
+* Sorted longest-first at build time.
+*/
+const TITLE_PREFIXES = [
+	"Ing.",
+	"Mgr.",
+	"MgA.",
+	"Bc.",
+	"BcA.",
+	"JUDr.",
+	"MUDr.",
+	"MVDr.",
+	"MDDr.",
+	"PhDr.",
+	"RNDr.",
+	"PaedDr.",
+	"ThDr.",
+	"ThLic.",
+	"ICDr.",
+	"RSDr.",
+	"PharmDr.",
+	"artD.",
+	"akad.",
+	"doc.",
+	"prof.",
+	"ao. Univ.-Prof.",
+	"o. Univ.-Prof.",
+	"Univ.-Prof.",
+	"Hon.-Prof.",
+	"em. Prof.",
+	"Dr. med. dent.",
+	"Dr. med. vet.",
+	"Dr. med.",
+	"Dr. rer. nat.",
+	"Dr. rer. soc.",
+	"Dr. rer. pol.",
+	"Dr. sc. tech.",
+	"Dr. sc. nat.",
+	"Dr. sc. hum.",
+	"Dr. iur.",
+	"Dr. jur.",
+	"Dr. theol.",
+	"Dr. oec.",
+	"Dr. techn.",
+	"Dr. h. c.",
+	"Dr. phil.",
+	"Dr.-Ing.",
+	"Dr. Ing.",
+	"Dr.",
+	"Dipl.-Wirt.-Ing.",
+	"Dipl.-Betriebsw.",
+	"Dipl.-Inform.",
+	"Dipl.-Volksw.",
+	"Dipl.-Psych.",
+	"Dipl.-Phys.",
+	"Dipl.-Chem.",
+	"Dipl.-Biol.",
+	"Dipl.-Math.",
+	"Dipl.-Päd.",
+	"Dipl.-Soz.",
+	"Dipl.-Kfm.",
+	"Dipl.-Jur.",
+	"Dipl. Ing.",
+	"Dipl.-Ing.",
+	"Mag. rer. soc. oec.",
+	"Mag. rer. nat.",
+	"Mag. phil.",
+	"Mag. iur.",
+	"Mag. arch.",
+	"Mag. pharm.",
+	"Mag. (FH)",
+	"Mag.",
+	"Bakk. rer. nat.",
+	"Bakk. techn.",
+	"Bakk. phil.",
+	"Bakk.",
+	"Lic. phil.",
+	"Lic. iur.",
+	"Lic. oec.",
+	"Lic. theol.",
+	"Lic.",
+	"Priv.-Doz.",
+	"PD",
+	"RA"
+];
+/**
+* Courtesy/honorific titles that precede a person's
+* name. Sorted alphabetically. The detector escapes
+* dots and adds \b for entries in HONORIFIC_BOUNDARY.
+*/
+const HONORIFICS = [
+	"Avv.",
+	"Dame",
+	"Doamna",
+	"Domnul",
+	"Don",
+	"Doña",
+	"Dott.",
+	"Judge",
+	"Justice",
+	"Lady",
+	"Lord",
+	"M.",
+	"Madame",
+	"Mademoiselle",
+	"Maître",
+	"Me",
+	"Messrs",
+	"Miss",
+	"Mlle",
+	"Mme",
+	"Monsieur",
+	"Mr",
+	"Mrs",
+	"Ms",
+	"Pr",
+	"Pr.",
+	"President",
+	"Señor",
+	"Señora",
+	"Sig.",
+	"Sig.ra",
+	"Signor",
+	"Signora",
+	"Signorina",
+	"Sir",
+	"Sr.",
+	"Sra."
+];
+/**
+* Honorifics that need \b word-boundary anchors
+* (short or common words that could match mid-word).
+*/
+const HONORIFIC_BOUNDARY = new Set([
+	"Don",
+	"Doña",
+	"M.",
+	"Me",
+	"Pr",
+	"Pr.",
+	"Señor",
+	"Señora"
+]);
+/**
+* Honorifics that are abbreviations: a dot after them is an
+* abbreviation dot (same sentence), so the detector keeps an
+* optional `.` between the title and the name ("Mr. Smith").
+* Every other honorific is a full word; a trailing dot ends a
+* sentence, so the detector must NOT consume it (otherwise a span
+* like "President. The Employee" crosses the sentence boundary).
+*
+* Maintained explicitly, NOT derived from "ends in a dot": "Mr" is
+* an abbreviation written without a dot, and "Lord" is a full word.
+*/
+const HONORIFIC_ABBREVIATION = new Set([
+	"Avv.",
+	"Dott.",
+	"M.",
+	"Me",
+	"Messrs",
+	"Mlle",
+	"Mme",
+	"Mr",
+	"Mrs",
+	"Ms",
+	"Pr",
+	"Pr.",
+	"Sig.",
+	"Sig.ra",
+	"Sr.",
+	"Sra."
+]);
+/**
+* Post-nominal degrees (comma or space separated after name).
+* Plain text; the detector auto-escapes for regex.
+*/
+const POST_NOMINALS = [
+	"Ph.D.",
+	"Ph.D",
+	"CSc.",
+	"DrSc.",
+	"ArtD.",
+	"D.Phil.",
+	"DPhil.",
+	"MPhil.",
+	"MBA",
+	"MPA",
+	"LL.M.",
+	"LL.B.",
+	"M.Sc.",
+	"B.Sc.",
+	"MSc.",
+	"BSc.",
+	"M.Eng.",
+	"B.Eng.",
+	"M.A.",
+	"B.A.",
+	"JCD",
+	"JD",
+	"DiS.",
+	"ACCA",
+	"FCCA",
+	"CIPM",
+	"CIPT",
+	"CIPP/E",
+	"CIPP",
+	"KC",
+	"QC"
+];
+const NONWESTERN_HONORIFICS = {
+	in: [
+		"Sri",
+		"Shri",
+		"Smt",
+		"Smt.",
+		"Kumari",
+		"Pandit",
+		"Pt.",
+		"Adv.",
+		"Adv",
+		"Justice",
+		"Hon'ble"
+	],
+	ar: [
+		"Al-",
+		"El-",
+		"Sheikh",
+		"Shaikh",
+		"Sheikha",
+		"Ustaz",
+		"Ustaza",
+		"Abu",
+		"Umm"
+	],
+	"zh-Latn": [
+		"Encik",
+		"Puan",
+		"Datuk",
+		"Dato'",
+		"Dato",
+		"Tan Sri",
+		"Tun"
+	],
+	"ja-Latn": [
+		"-san",
+		"-sama",
+		"-sensei",
+		"Mr.",
+		"Ms."
+	],
+	ko: ["Sunsaeng", "Gyosu"],
+	th: [
+		"Khun",
+		"Nai",
+		"Nang",
+		"Khunying",
+		"Luang",
+		"Phra",
+		"Mom",
+		"Momluang",
+		"Mommuen",
+		"Momratchawong"
+	],
+	vi: [
+		"Ông",
+		"Ong",
+		"Ba",
+		"Co"
+	],
+	fil: [
+		"Atty.",
+		"Atty",
+		"Ginoo",
+		"Ginang",
+		"Binibini",
+		"G.",
+		"Gng.",
+		"Bb."
+	],
+	id: [
+		"Bapak",
+		"Ibu",
+		"Pak",
+		"Bu",
+		"Raden",
+		"Tuan",
+		"Nyonya",
+		"Nona",
+		"Haji",
+		"Hajjah",
+		"H.",
+		"Hj.",
+		"S.H.",
+		"S.H",
+		"Ir.",
+		"Drs."
+	]
+};
+//#endregion
 //#region src/util/text.ts
 /**
 * Shared text utilities for detectors.
@@ -7985,6 +8292,30 @@ const getCorpus = (ctx) => ctx.nameCorpus;
 const getNameCorpusFirstNames = (ctx = defaultContext) => ctx.nameCorpus?.firstNamesList ?? [];
 const getNameCorpusSurnames = (ctx = defaultContext) => ctx.nameCorpus?.surnamesList ?? [];
 const getNameCorpusTitles = (ctx = defaultContext) => ctx.nameCorpus?.titlesList ?? [];
+const getNameCorpusNonWesternNames = (ctx = defaultContext) => ctx.nameCorpus?.nonWesternNamesList ?? [];
+const NONWESTERN_LOCALE_KEYS = [
+	"in",
+	"ar",
+	"ja-latn",
+	"ko",
+	"zh-latn",
+	"th",
+	"vi",
+	"fil",
+	"id"
+];
+const normalizeCorpusLanguage = (language) => language.toLowerCase();
+const getScopedNonWesternLocaleKeys = (languages) => {
+	if (languages === void 0) return NONWESTERN_LOCALE_KEYS;
+	const allowed = new Set(languages.map(normalizeCorpusLanguage));
+	return NONWESTERN_LOCALE_KEYS.filter((locale) => allowed.has(locale));
+};
+const getScopedNonWesternHonorifics = (languages) => {
+	const entries = Object.entries(NONWESTERN_HONORIFICS);
+	if (languages === void 0) return entries.flatMap(([, forms]) => forms);
+	const allowed = new Set(languages.map(normalizeCorpusLanguage));
+	return entries.filter(([locale]) => allowed.has(normalizeCorpusLanguage(locale))).flatMap(([, forms]) => forms);
+};
 /**
 * Load name corpus data from injected dictionaries
 * and legacy config files. Merges all sources.
@@ -8035,17 +8366,42 @@ const initNameCorpus = (ctx = defaultContext, dictionaries, languages) => {
 			const commonWords = new Set(commonWordsMod.default.words.map((word) => word.toLowerCase()));
 			const dedupFirst = dedup(firstNames);
 			const dedupSurnames = dedup(surnames).filter((name) => !commonWords.has(name.toLowerCase()));
-			const titles = titleMod.default.tokens;
+			const titles = [...titleMod.default.tokens];
+			const scopedNonWesternHonorifics = getScopedNonWesternHonorifics(languages);
+			for (const form of scopedNonWesternHonorifics) titles.push(form.replace(/[.-]+$/u, "").toLowerCase());
+			const dedupTitles = dedup(titles);
+			const titleAbbrSet = /* @__PURE__ */ new Set();
+			for (const prefix of TITLE_PREFIXES) if (prefix.endsWith(".")) titleAbbrSet.add(prefix.replace(/[.-]+$/u, "").toLowerCase());
+			for (const form of HONORIFIC_ABBREVIATION) titleAbbrSet.add(form.replace(/[.-]+$/u, "").toLowerCase());
+			for (const form of scopedNonWesternHonorifics) if (form.endsWith(".")) titleAbbrSet.add(form.replace(/[.-]+$/u, "").toLowerCase());
 			const exclusions = exclusionMod.default.words;
+			const nwLocaleKeys = getScopedNonWesternLocaleKeys(languages);
+			const [nwNameMods, nwExcludedMod] = await Promise.all([Promise.all(nwLocaleKeys.map((locale) => import(`../data/names-nw-${locale}.json`))), import("./names-nw-excluded-allcaps.mjs")]);
+			const nonWesternNames = [];
+			for (const mod of nwNameMods) for (const name of mod.default.names) nonWesternNames.push(name);
+			if (dictionaries?.nonWesternNames) {
+				const entries = languages === void 0 ? Object.entries(dictionaries.nonWesternNames) : (() => {
+					const allowed = new Set(languages.map(normalizeCorpusLanguage));
+					return Object.entries(dictionaries.nonWesternNames).filter(([language]) => allowed.has(normalizeCorpusLanguage(language)));
+				})();
+				for (const [, names] of entries) for (const name of names) nonWesternNames.push(name);
+			}
+			const dedupNonWestern = dedup(nonWesternNames);
+			const dedupExcludedAllCaps = dedup(nwExcludedMod.default.words);
 			ctx.nameCorpus = {
 				firstNames: Object.freeze(new Set(dedupFirst)),
 				surnames: Object.freeze(new Set(dedupSurnames)),
-				titleTokens: Object.freeze(new Set(titles)),
+				titleTokens: Object.freeze(new Set(dedupTitles)),
+				titleAbbreviations: Object.freeze(titleAbbrSet),
 				excludedWords: Object.freeze(new Set(exclusions)),
+				nonWesternNames: Object.freeze(new Set(dedupNonWestern)),
+				excludedAllCaps: Object.freeze(new Set(dedupExcludedAllCaps)),
 				firstNamesList: Object.freeze(dedupFirst),
 				surnamesList: Object.freeze(dedupSurnames),
-				titlesList: Object.freeze(titles),
-				excludedList: Object.freeze(exclusions)
+				titlesList: Object.freeze(dedupTitles),
+				excludedList: Object.freeze(exclusions),
+				nonWesternNamesList: Object.freeze(dedupNonWestern),
+				excludedAllCapsList: Object.freeze(dedupExcludedAllCaps)
 			};
 		} catch (err) {
 			ctx.nameCorpusPromise = null;
@@ -8055,51 +8411,237 @@ const initNameCorpus = (ctx = defaultContext, dictionaries, languages) => {
 	ctx.nameCorpusPromise = promise;
 	return promise;
 };
-const INFLECTION_SUFFIXES = [
-	"ovi",
-	"em",
-	"om",
-	"ou",
-	"é",
-	"a",
-	"u"
-];
-/**
-* Strip common Czech/Slovak case suffixes from a token.
-* Returns candidate base forms if stripping produces a
-* plausible name (capitalised, length >= 3).
-*
-* For the "-ou" instrumental feminine suffix, also yields
-* base + "a" (e.g., "Editou" → "Edit" and "Edita")
-* because Czech feminine names decline -a → -ou.
-*/
-const stripInflection = (token) => {
-	const candidates = [];
-	for (const suffix of INFLECTION_SUFFIXES) if (token.length > suffix.length + 2 && token.endsWith(suffix)) {
-		const base = token.slice(0, -suffix.length);
-		if (/^\p{Lu}/u.test(base)) {
-			candidates.push(base);
-			if (suffix === "ou" || suffix === "é" || suffix === "u") candidates.push(`${base}a`);
-		}
-	}
-	return candidates;
-};
-const TOKEN_TYPE = {
-	NAME: "name",
-	SURNAME: "surname",
-	TITLE: "title",
-	ABBREVIATION: "abbreviation",
-	CAPITALIZED: "capitalized",
-	OTHER: "other"
-};
-const PERSON_CHAIN_BREAK_RE$1 = /[!?;:]/u;
-const isInitialContinuationGap$1 = (text, gap) => /^\p{Lu}$/u.test(text) && /^\.[^\S\n]{1,2}$/u.test(gap) || /^[^\S\n]{1,2}(?:\p{Lu}\.[^\S\n]{1,2})+$/u.test(gap);
-/**
-* Check if a token is in the first-name set, either
-* directly or after stripping Czech/Slovak inflection.
-*/
-const isFirstNameToken = (token, corpus) => {
-	if (corpus.firstNames.has(token)) return true;
+const DECLENSION_RULES = [
+	{
+		ending: "",
+		gate: /[bdfghklmnpqrstvwxzł]$/u,
+		forms: [
+			"a",
+			"u",
+			"ovi",
+			"em",
+			"om"
+		]
+	},
+	{
+		ending: "",
+		gate: /[bdflmnpstvwz]$/u,
+		forms: ["e"]
+	},
+	{
+		ending: "",
+		gate: /[cčďťňřšžjľ]$/u,
+		forms: [
+			"e",
+			"i",
+			"a",
+			"ovi",
+			"em",
+			"om"
+		]
+	},
+	{
+		ending: "ek",
+		gate: /[^aeiouyáéěíóôúůý]ek$/u,
+		forms: [
+			"ka",
+			"ku",
+			"kovi",
+			"kem",
+			"kom"
+		]
+	},
+	{
+		ending: "el",
+		gate: /[^aeiouyáéěíóôúůý]el$/u,
+		forms: [
+			"la",
+			"lu",
+			"le",
+			"lovi",
+			"lem",
+			"lom"
+		]
+	},
+	{
+		ending: "ec",
+		gate: /[^aeiouyáéěíóôúůý]ec$/u,
+		forms: [
+			"ce",
+			"ci",
+			"covi",
+			"cem",
+			"com"
+		]
+	},
+	{
+		ending: "a",
+		gate: /[^cčďťňřšžji]a$/u,
+		forms: [
+			"y",
+			"u",
+			"o",
+			"ou",
+			"ovi"
+		]
+	},
+	{
+		ending: "a",
+		gate: /[cčďťňřšžj]a$/u,
+		forms: [
+			"i",
+			"u",
+			"o",
+			"ou",
+			"ovi"
+		]
+	},
+	{
+		ending: "a",
+		gate: /ia$/u,
+		forms: [
+			"e",
+			"i",
+			"u",
+			"ou"
+		]
+	},
+	{
+		ending: "ka",
+		gate: /ka$/u,
+		forms: ["ce"]
+	},
+	{
+		ending: "ra",
+		gate: /ra$/u,
+		forms: ["ře", "re"]
+	},
+	{
+		ending: "ha",
+		gate: /ha$/u,
+		forms: ["ze"]
+	},
+	{
+		ending: "ga",
+		gate: /ga$/u,
+		forms: ["ze"]
+	},
+	{
+		ending: "cha",
+		gate: /cha$/u,
+		forms: ["še"]
+	},
+	{
+		ending: "a",
+		gate: /[bdfmnptv]a$/u,
+		forms: ["ě", "e"]
+	},
+	{
+		ending: "a",
+		gate: /[szl]a$/u,
+		forms: ["e"]
+	},
+	{
+		ending: "á",
+		gate: /á$/u,
+		forms: [
+			"é",
+			"ou",
+			"ej",
+			"ú"
+		]
+	},
+	{
+		ending: "ý",
+		gate: /ý$/u,
+		forms: [
+			"ého",
+			"ému",
+			"ém",
+			"ým"
+		]
+	},
+	{
+		ending: "",
+		gate: /[íiy]$/u,
+		forms: [
+			"ho",
+			"mu",
+			"m"
+		]
+	},
+	{
+		ending: "e",
+		gate: /[^aeouyáéěíóôúůý]e$/u,
+		forms: ["i", "í"]
+	},
+	{
+		ending: "o",
+		gate: /o$/u,
+		forms: [
+			"a",
+			"ovi",
+			"em",
+			"om"
+		]
+	}
+];
+/**
+* Generate declined Czech/Slovak variants of a
+* nominative name. Only variants licensed by the
+* ending-shape rules are produced, which bounds the
+* deny-list AC pattern growth. Returns an empty array
+* for names too short to decline safely.
+*/
+const expandNameDeclensions = (name) => {
+	if (name.length < 3) return [];
+	const lower = name.toLowerCase();
+	const variants = [];
+	for (const rule of DECLENSION_RULES) {
+		if (!rule.gate.test(lower)) continue;
+		const stem = name.slice(0, name.length - rule.ending.length);
+		if (stem.length < 2) continue;
+		for (const form of rule.forms) variants.push(stem + form);
+	}
+	return variants;
+};
+/**
+* Strip Czech/Slovak case endings from a token using
+* the inverse of DECLENSION_RULES. Returns candidate
+* nominative forms; each candidate is validated against
+* the rule's shape gate, so only bases the paradigm
+* could actually have declined are proposed. Callers
+* check candidates against the name corpus.
+*/
+const stripInflection = (token) => {
+	const candidates = [];
+	for (const rule of DECLENSION_RULES) for (const form of rule.forms) {
+		if (token.length <= form.length + 2 || !token.endsWith(form)) continue;
+		const base = token.slice(0, token.length - form.length) + rule.ending;
+		if (!/^\p{Lu}/u.test(base)) continue;
+		if (!rule.gate.test(base.toLowerCase())) continue;
+		candidates.push(base);
+	}
+	return candidates;
+};
+const TOKEN_TYPE = {
+	NAME: "name",
+	SURNAME: "surname",
+	TITLE: "title",
+	ABBREVIATION: "abbreviation",
+	JA_SUFFIX: "ja_suffix",
+	ARABIC_CONNECTOR: "arabic_connector",
+	CAPITALIZED: "capitalized",
+	OTHER: "other"
+};
+const PERSON_CHAIN_BREAK_RE$1 = /[!?;:]/u;
+const isInitialContinuationGap$1 = (text, gap) => /^\p{Lu}$/u.test(text) && /^\.[^\S\n]{1,2}$/u.test(gap) || /^[^\S\n]{1,2}(?:\p{Lu}\.[^\S\n]{1,2})+$/u.test(gap);
+/**
+* Check if a token is in the first-name set, either
+* directly or after stripping Czech/Slovak inflection.
+*/
+const isFirstNameToken = (token, corpus) => {
+	if (corpus.firstNames.has(token)) return true;
 	return stripInflection(token).some((b) => corpus.firstNames.has(b));
 };
 /**
@@ -8115,6 +8657,141 @@ const isSurnameToken = (token, corpus) => {
 * abbreviation: "J.", "M.", etc.
 */
 const isAbbreviation = (token) => token.length === 2 && /^\p{Lu}$/u.test(token[0] ?? "") && token[1] === ".";
+/**
+* Title-case a token, re-capitalizing after apostrophes
+* so D'Souza, O'Brien, Hon'ble match the corpus.
+*/
+const titleCaseWithApostrophe = (text) => (text[0]?.toUpperCase() ?? "") + text.slice(1).toLowerCase().replace(/'\p{Ll}/gu, (m) => m.toUpperCase());
+/**
+* Check if a token is in the non-Western name corpus.
+* Tries both title-cased (with apostrophe re-cap) and
+* raw forms.
+*/
+const isNonWesternNameToken = (token, corpus) => {
+	if (corpus.nonWesternNames.has(token)) return true;
+	return corpus.nonWesternNames.has(titleCaseWithApostrophe(token));
+};
+/** Japanese honorific suffixes attached via hyphen. */
+const JA_SUFFIXES = new Set([
+	"san",
+	"sama",
+	"sensei"
+]);
+/** Arabic patronymic connectors. */
+const ARABIC_CONNECTORS = new Set([
+	"bin",
+	"bint",
+	"ibn",
+	"al",
+	"el"
+]);
+/** Al-/El- prefixed name pattern (e.g., Al-Rashid, El-Amin). */
+const AL_EL_PREFIX_RE = /^[Aa]l-[A-Z][a-z]+$|^[Ee]l-[A-Z][a-z]+$/u;
+/** CJK name detection: 2-4 Han chars not adjacent to others. */
+const CJK_NAME_RE = /(?<!\p{Script=Han})\p{Script=Han}{2,4}(?!\p{Script=Han})/gu;
+/** Han character threshold for CJK-majority documents. */
+const CJK_HAN_RATIO = .15;
+const CJK_NON_PERSON_TERMS = new Set([
+	"中国",
+	"中國",
+	"中文",
+	"人民",
+	"公司",
+	"香港",
+	"台湾",
+	"臺灣",
+	"日本",
+	"韩国",
+	"韓國"
+]);
+const CJK_SURNAME_CHARS = new Set([
+	"王",
+	"李",
+	"张",
+	"張",
+	"刘",
+	"劉",
+	"陈",
+	"陳",
+	"杨",
+	"楊",
+	"黄",
+	"黃",
+	"赵",
+	"趙",
+	"吴",
+	"吳",
+	"周",
+	"徐",
+	"孙",
+	"孫",
+	"马",
+	"馬",
+	"朱",
+	"胡",
+	"郭",
+	"何",
+	"林",
+	"高",
+	"梁",
+	"郑",
+	"鄭",
+	"罗",
+	"羅",
+	"宋",
+	"谢",
+	"謝",
+	"唐",
+	"韩",
+	"韓",
+	"曹",
+	"许",
+	"許",
+	"邓",
+	"鄧",
+	"萧",
+	"蕭",
+	"田",
+	"山",
+	"佐",
+	"鈴",
+	"渡",
+	"伊",
+	"中",
+	"小",
+	"吉",
+	"金",
+	"朴",
+	"박",
+	"김",
+	"이",
+	"최",
+	"정",
+	"강",
+	"조",
+	"윤",
+	"장",
+	"임",
+	"한"
+].join(""));
+const isLikelyCjkPersonName = (text) => {
+	if (CJK_NON_PERSON_TERMS.has(text)) return false;
+	const first = text.at(0);
+	return first !== void 0 && CJK_SURNAME_CHARS.has(first);
+};
+/** Organization keyword filter. */
+const ORG_WORDS = /\b(?:Group|Company|LLC|LLP|LP|Inc|Ltd|Corp|Corporation|Holdings|Partners|Association|University|Bank|Fund|Trust|Agency|Government|Ministry|Office|Department|Council|Board|Committee|Commission|Services|Solutions|Technologies|Systems|Analytics|Software)\b/i;
+const isOrganization = (text) => ORG_WORDS.test(text);
+/** Deduplicate overlapping entity spans (first wins). */
+const deduplicateSpans$1 = (entities) => {
+	const sorted = [...entities].sort((a, b) => a.start - b.start || b.end - a.end);
+	const result = [];
+	for (const entity of sorted) {
+		const last = result.at(-1);
+		if (!last || entity.start >= last.end) result.push(entity);
+	}
+	return result;
+};
 const segmenter$1 = new Intl.Segmenter(void 0, { granularity: "word" });
 /**
 * Split text into word segments using Intl.Segmenter.
@@ -8163,21 +8840,55 @@ const classifyToken = (word, corpus, fullText) => {
 		text,
 		type: TOKEN_TYPE.TITLE,
 		start,
+		end,
+		...corpus.titleAbbreviations.has(stripped) ? { titleAbbreviation: true } : {}
+	};
+	if (JA_SUFFIXES.has(lower)) return {
+		text,
+		type: TOKEN_TYPE.JA_SUFFIX,
+		start,
 		end
 	};
+	if (ARABIC_CONNECTORS.has(lower)) return {
+		text,
+		type: TOKEN_TYPE.ARABIC_CONNECTOR,
+		start,
+		end
+	};
+	if (AL_EL_PREFIX_RE.test(text)) return {
+		text,
+		type: TOKEN_TYPE.NAME,
+		start,
+		end,
+		nonWestern: true
+	};
 	if (isAbbreviation(text)) return {
 		text,
 		type: TOKEN_TYPE.ABBREVIATION,
 		start,
 		end
 	};
+	if (/^(?:\p{Lu}\.)+\p{Lu}?$/u.test(text)) return {
+		text,
+		type: TOKEN_TYPE.ABBREVIATION,
+		start,
+		end
+	};
 	if (text.length === 1 && UPPER_START_RE.test(text) && fullText[end] === ".") {
 		const lineStart = fullText.lastIndexOf("\n", start - 1) + 1;
 		const before = fullText.slice(lineStart, start).trimEnd();
 		const lastWord = /\p{L}[\p{L}\p{M}'-]*$/u.exec(before)?.[0];
-		if (lastWord) {
-			const lookup = (token) => isFirstNameToken(token, corpus) || isFirstNameToken((token[0] ?? "") + token.slice(1).toLowerCase(), corpus);
-			if (lookup(lastWord)) return {
+		const lookup = (token) => isFirstNameToken(token, corpus) || isFirstNameToken((token[0] ?? "") + token.slice(1).toLowerCase(), corpus) || isNonWesternNameToken(token, corpus);
+		if (lastWord && lookup(lastWord)) return {
+			text,
+			type: TOKEN_TYPE.ABBREVIATION,
+			start,
+			end
+		};
+		const afterDot = fullText.slice(end + 1).trimStart();
+		const nextWord = /^\p{L}[\p{L}\p{M}'-]*/u.exec(afterDot)?.[0];
+		if (nextWord) {
+			if (lookup(nextWord) || nextWord.length === 1 && UPPER_START_RE.test(nextWord)) return {
 				text,
 				type: TOKEN_TYPE.ABBREVIATION,
 				start,
@@ -8197,26 +8908,55 @@ const classifyToken = (word, corpus, fullText) => {
 		start,
 		end
 	};
-	if (text.length < 3) return {
+	if (text.length < 2) return {
+		text,
+		type: TOKEN_TYPE.OTHER,
+		start,
+		end
+	};
+	if (text.length < 3 && !isNonWesternNameToken(text, corpus) && !JA_SUFFIXES.has(lower) && !ARABIC_CONNECTORS.has(lower) && !(ALL_UPPER_RE.test(text) && !corpus.excludedAllCaps.has(text))) return {
 		text,
 		type: TOKEN_TYPE.OTHER,
 		start,
 		end
 	};
 	if (text.length >= 3 && ALL_UPPER_RE.test(text)) {
+		if (corpus.excludedAllCaps.has(text)) return {
+			text,
+			type: TOKEN_TYPE.OTHER,
+			start,
+			end
+		};
+		const titleCased = (text[0] ?? "") + text.slice(1).toLowerCase();
+		const nwMatch = isNonWesternNameToken(titleCased, corpus);
+		if (nwMatch && !isFirstNameToken(titleCased, corpus)) return {
+			text,
+			type: TOKEN_TYPE.NAME,
+			start,
+			end,
+			nonWestern: true
+		};
 		if (isAllCapsContextLine(fullText, start) && isAllCapsLineNameShaped(fullText, start)) {
-			const titleCased = (text[0] ?? "") + text.slice(1).toLowerCase();
 			if (isFirstNameToken(titleCased, corpus)) return {
 				text,
 				type: TOKEN_TYPE.NAME,
 				start,
-				end
+				end,
+				...nwMatch ? { nonWestern: true } : {}
 			};
 			if (isSurnameToken(titleCased, corpus)) return {
 				text,
 				type: TOKEN_TYPE.SURNAME,
 				start,
-				end
+				end,
+				...nwMatch ? { nonWestern: true } : {}
+			};
+			if (nwMatch) return {
+				text,
+				type: TOKEN_TYPE.NAME,
+				start,
+				end,
+				nonWestern: true
 			};
 		}
 		return {
@@ -8232,17 +8972,32 @@ const classifyToken = (word, corpus, fullText) => {
 		start,
 		end
 	};
-	if (isFirstNameToken(text, corpus)) return {
+	if (isFirstNameToken(text, corpus)) {
+		const nw = isNonWesternNameToken(text, corpus);
+		return {
+			text,
+			type: TOKEN_TYPE.NAME,
+			start,
+			end,
+			...nw ? { nonWestern: true } : {}
+		};
+	}
+	if (isSurnameToken(text, corpus)) {
+		const nw = isNonWesternNameToken(text, corpus);
+		return {
+			text,
+			type: TOKEN_TYPE.SURNAME,
+			start,
+			end,
+			...nw ? { nonWestern: true } : {}
+		};
+	}
+	if (isNonWesternNameToken(text, corpus)) return {
 		text,
 		type: TOKEN_TYPE.NAME,
 		start,
-		end
-	};
-	if (isSurnameToken(text, corpus)) return {
-		text,
-		type: TOKEN_TYPE.SURNAME,
-		start,
-		end
+		end,
+		nonWestern: true
 	};
 	return {
 		text,
@@ -8254,11 +9009,12 @@ const classifyToken = (word, corpus, fullText) => {
 /**
 * Detect person names by looking up tokens against the
 * name corpus, then chaining adjacent name-like tokens.
+* Handles both Western and non-Western name patterns.
 *
 * Requires initNameCorpus() to have been called first.
 * If not initialized, returns an empty array.
 *
-* Scoring:
+* Scoring (Western):
 *   TITLE + NAME/SURNAME       → 0.95
 *   NAME + NAME/SURNAME        → 0.9
 *   SURNAME + NAME/SURNAME     → 0.9
@@ -8266,18 +9022,63 @@ const classifyToken = (word, corpus, fullText) => {
 *   ABBREVIATION + NAME        → 0.7
 *   Standalone NAME            → 0.5 (low confidence)
 *   Standalone SURNAME         → skip (too ambiguous)
-*/
-const detectNameCorpus = (fullText, ctx = defaultContext) => {
+*
+* Scoring (non-Western, when chain contains nonWestern tokens):
+*   TITLE + (nonWestern|CAPITALIZED) → 0.95
+*   JA_SUFFIX + (CAPITALIZED|nonWestern) → 0.9
+*   ARABIC_CONNECTOR + nonWestern     → 0.9
+*   2+ nonWestern tokens              → 0.9
+*   nonWestern + (CAPITALIZED|ABBREVIATION) → 0.9
+*   Standalone nonWestern mid-sentence → 0.5
+*/
+const detectNameCorpus = (fullText, ctx = defaultContext, options = {}) => {
 	const corpus = getCorpus(ctx);
 	if (!corpus) return [];
-	const tokens = segmentWords(fullText).map((w) => classifyToken(w, corpus, fullText));
+	const supplementalMode = options.mode === "supplemental";
 	const entities = [];
+	const threshold = Math.ceil(fullText.length * CJK_HAN_RATIO);
+	let hanCount = 0;
+	for (const _ of fullText.matchAll(/\p{Script=Han}/gu)) {
+		hanCount++;
+		if (hanCount >= threshold) break;
+	}
+	if (fullText.length < 100 || hanCount < threshold) {
+		CJK_NAME_RE.lastIndex = 0;
+		let match;
+		while ((match = CJK_NAME_RE.exec(fullText)) !== null) {
+			const cjkText = match[0];
+			if (isLikelyCjkPersonName(cjkText) && !isOrganization(cjkText)) entities.push({
+				start: match.index,
+				end: match.index + cjkText.length,
+				label: "person",
+				text: cjkText,
+				score: .95,
+				source: DETECTION_SOURCES.REGEX
+			});
+		}
+	}
+	const words = segmentWords(fullText);
+	const tokens = [];
+	for (let idx = 0; idx < words.length; idx++) {
+		const word = words[idx];
+		const lower = word.text.toLowerCase();
+		const nextWord = words[idx + 1];
+		if ((lower === "s" || lower === "d" || lower === "w" || lower === "r") && fullText[word.end] === "/" && nextWord && nextWord.start === word.end + 1 && nextWord.text.toLowerCase() === "o") {
+			tokens.push({
+				text: fullText.slice(word.start, word.end + 2),
+				type: TOKEN_TYPE.ARABIC_CONNECTOR,
+				start: word.start,
+				end: word.end + 2
+			});
+			idx++;
+		} else tokens.push(classifyToken(word, corpus, fullText));
+	}
 	const consumed = /* @__PURE__ */ new Set();
 	for (let i = 0; i < tokens.length; i++) {
 		if (consumed.has(i)) continue;
 		const token = tokens[i];
 		if (!token) continue;
-		if (token.type !== TOKEN_TYPE.TITLE && token.type !== TOKEN_TYPE.NAME && token.type !== TOKEN_TYPE.SURNAME && token.type !== TOKEN_TYPE.ABBREVIATION) continue;
+		if (token.type !== TOKEN_TYPE.TITLE && token.type !== TOKEN_TYPE.NAME && token.type !== TOKEN_TYPE.SURNAME && token.type !== TOKEN_TYPE.ABBREVIATION && token.type !== TOKEN_TYPE.ARABIC_CONNECTOR) continue;
 		const MAX_CHAIN = 5;
 		const chain = [token];
 		let j = i + 1;
@@ -8287,10 +9088,12 @@ const detectNameCorpus = (fullText, ctx = defaultContext) => {
 			const prev = chain.at(-1);
 			if (prev) {
 				const gap = fullText.slice(prev.end, next.start);
-				const breaksOnPeriod = gap.includes(".") && !isInitialContinuationGap$1(prev.text, gap);
+				const periodIsPartOfPrevToken = prev.type === TOKEN_TYPE.ABBREVIATION || prev.type === TOKEN_TYPE.TITLE && prev.titleAbbreviation === true;
+				const breaksOnPeriod = gap.includes(".") && !isInitialContinuationGap$1(prev.text, gap) && !periodIsPartOfPrevToken;
 				if (gap.includes("\n") || PERSON_CHAIN_BREAK_RE$1.test(gap) || breaksOnPeriod) break;
+				if (next.type === TOKEN_TYPE.JA_SUFFIX && gap !== "-" && gap.trim() !== "") break;
 			}
-			if (next.type === TOKEN_TYPE.NAME || next.type === TOKEN_TYPE.SURNAME || next.type === TOKEN_TYPE.TITLE || next.type === TOKEN_TYPE.ABBREVIATION || next.type === TOKEN_TYPE.CAPITALIZED) {
+			if (next.type !== TOKEN_TYPE.OTHER) {
 				chain.push(next);
 				j++;
 			} else break;
@@ -8299,10 +9102,24 @@ const detectNameCorpus = (fullText, ctx = defaultContext) => {
 		const hasCorpusName = chain.some((t) => isCorpusMatch(t.type));
 		const hasFirstName = chain.some((t) => t.type === TOKEN_TYPE.NAME);
 		const hasAbbreviation = chain.some((t) => t.type === TOKEN_TYPE.ABBREVIATION);
+		const hasNonWestern = chain.some((t) => t.nonWestern === true);
+		const hasJaSuffix = chain.some((t) => t.type === TOKEN_TYPE.JA_SUFFIX);
+		const hasArabicConnector = chain.some((t) => t.type === TOKEN_TYPE.ARABIC_CONNECTOR);
 		const corpusCount = chain.filter((t) => isCorpusMatch(t.type)).length;
 		const capitalizedCount = chain.filter((t) => t.type === TOKEN_TYPE.CAPITALIZED).length;
+		const nonWesternCount = chain.filter((t) => t.nonWestern).length;
 		let score;
-		if (hasTitle && hasCorpusName) score = .95;
+		if (hasNonWestern) {
+			if (hasTitle && (nonWesternCount > 0 || capitalizedCount > 0)) score = .95;
+			else if (hasJaSuffix && (capitalizedCount > 0 || nonWesternCount > 0)) score = .9;
+			else if (hasArabicConnector && nonWesternCount > 0) score = .9;
+			else if (nonWesternCount >= 2) score = .9;
+			else if (nonWesternCount > 0 && (capitalizedCount > 0 || hasAbbreviation)) score = .9;
+			else if (nonWesternCount === 1 && chain.length === 1 && !isSentenceStart(fullText, token.start)) score = .5;
+			else continue;
+			if (supplementalMode && score < .9) continue;
+		} else if (supplementalMode) continue;
+		else if (hasTitle && hasCorpusName) score = .95;
 		else if (corpusCount >= 2) score = .9;
 		else if (hasCorpusName && capitalizedCount > 0) score = .7;
 		else if (hasAbbreviation && hasCorpusName) score = .7;
@@ -8313,7 +9130,10 @@ const detectNameCorpus = (fullText, ctx = defaultContext) => {
 			score = .5;
 		} else if (!hasFirstName && chain.length === 1 && chain[0]?.type === TOKEN_TYPE.SURNAME) continue;
 		else if (hasTitle && chain.length === 1) continue;
-		else {
+		else if (hasJaSuffix || hasArabicConnector) {
+			if (!hasCorpusName && !hasFirstName) continue;
+			score = .5;
+		} else {
 			if (!hasCorpusName) continue;
 			score = .5;
 		}
@@ -8323,6 +9143,7 @@ const detectNameCorpus = (fullText, ctx = defaultContext) => {
 		const start = first.start;
 		const end = last.end;
 		const text = fullText.slice(start, end);
+		if (isOrganization(text)) continue;
 		for (let k = i; k < i + chain.length; k++) consumed.add(k);
 		entities.push({
 			start,
@@ -8333,14 +9154,14 @@ const detectNameCorpus = (fullText, ctx = defaultContext) => {
 			source: DETECTION_SOURCES.REGEX
 		});
 	}
-	return entities;
+	return deduplicateSpans$1(entities);
 };
 //#endregion
 //#region src/detectors/signatures.ts
 const SLASH_S_RE = /\/s\/[ \t]*/g;
 const LABELLED_NAME_RE = /\b(?:By|Name)[ \t]*:[ \t]*(?:\/s\/[ \t]+)?([^\n]*?)(?=\s{3,}|[\t]|\n|$)/gi;
 const WITNESS_ANCHOR_RE = /\bIN WITNESS WHEREOF\b[^\n]*\n/gi;
-const NAME_PARTICLE = "(?:de|del|della|der|den|di|du|el|la|le|van|von|y|zu|af|ben|bin|al|d'|d’)";
+const NAME_PARTICLE = "(?:de|del|della|der|den|di|du|da|das|do|dos|el|la|le|van|von|y|zu|af|ben|bin|al|d'|d’)";
 const CAP_TOKEN = "\\p{Lu}[\\p{L}\\p{M}.'\\-]{0,30}";
 const NAME_SHAPE_RE = new RegExp(`^${CAP_TOKEN}(?:[ \\t]+(?:${NAME_PARTICLE}|${CAP_TOKEN})){1,4}$`, "u");
 const MAX_NAME_LEN = 60;
@@ -8445,219 +9266,6 @@ const detectSignatures = (fullText, _ctx = defaultContext) => {
 	return results;
 };
 //#endregion
-//#region src/config/titles.ts
-/**
-* Academic and professional title prefixes.
-* Plain text; the detector auto-escapes for regex.
-* Sorted longest-first at build time.
-*/
-const TITLE_PREFIXES = [
-	"Ing.",
-	"Mgr.",
-	"MgA.",
-	"Bc.",
-	"BcA.",
-	"JUDr.",
-	"MUDr.",
-	"MVDr.",
-	"MDDr.",
-	"PhDr.",
-	"RNDr.",
-	"PaedDr.",
-	"ThDr.",
-	"ThLic.",
-	"ICDr.",
-	"RSDr.",
-	"PharmDr.",
-	"artD.",
-	"akad.",
-	"doc.",
-	"prof.",
-	"ao. Univ.-Prof.",
-	"o. Univ.-Prof.",
-	"Univ.-Prof.",
-	"Hon.-Prof.",
-	"em. Prof.",
-	"Dr. med. dent.",
-	"Dr. med. vet.",
-	"Dr. med.",
-	"Dr. rer. nat.",
-	"Dr. rer. soc.",
-	"Dr. rer. pol.",
-	"Dr. sc. tech.",
-	"Dr. sc. nat.",
-	"Dr. sc. hum.",
-	"Dr. iur.",
-	"Dr. jur.",
-	"Dr. theol.",
-	"Dr. oec.",
-	"Dr. techn.",
-	"Dr. h. c.",
-	"Dr. phil.",
-	"Dr.-Ing.",
-	"Dr. Ing.",
-	"Dr.",
-	"Dipl.-Wirt.-Ing.",
-	"Dipl.-Betriebsw.",
-	"Dipl.-Inform.",
-	"Dipl.-Volksw.",
-	"Dipl.-Psych.",
-	"Dipl.-Phys.",
-	"Dipl.-Chem.",
-	"Dipl.-Biol.",
-	"Dipl.-Math.",
-	"Dipl.-Päd.",
-	"Dipl.-Soz.",
-	"Dipl.-Kfm.",
-	"Dipl.-Jur.",
-	"Dipl. Ing.",
-	"Dipl.-Ing.",
-	"Mag. rer. soc. oec.",
-	"Mag. rer. nat.",
-	"Mag. phil.",
-	"Mag. iur.",
-	"Mag. arch.",
-	"Mag. pharm.",
-	"Mag. (FH)",
-	"Mag.",
-	"Bakk. rer. nat.",
-	"Bakk. techn.",
-	"Bakk. phil.",
-	"Bakk.",
-	"Lic. phil.",
-	"Lic. iur.",
-	"Lic. oec.",
-	"Lic. theol.",
-	"Lic.",
-	"Priv.-Doz.",
-	"PD",
-	"RA"
-];
-/**
-* Courtesy/honorific titles that precede a person's
-* name. Sorted alphabetically. The detector escapes
-* dots and adds \b for entries in HONORIFIC_BOUNDARY.
-*/
-const HONORIFICS = [
-	"Avv.",
-	"Dame",
-	"Doamna",
-	"Domnul",
-	"Don",
-	"Doña",
-	"Dott.",
-	"Judge",
-	"Justice",
-	"Lady",
-	"Lord",
-	"M.",
-	"Madame",
-	"Mademoiselle",
-	"Maître",
-	"Me",
-	"Messrs",
-	"Miss",
-	"Mlle",
-	"Mme",
-	"Monsieur",
-	"Mr",
-	"Mrs",
-	"Ms",
-	"Pr",
-	"Pr.",
-	"President",
-	"Señor",
-	"Señora",
-	"Sig.",
-	"Sig.ra",
-	"Signor",
-	"Signora",
-	"Signorina",
-	"Sir",
-	"Sr.",
-	"Sra."
-];
-/**
-* Honorifics that need \b word-boundary anchors
-* (short or common words that could match mid-word).
-*/
-const HONORIFIC_BOUNDARY = new Set([
-	"Don",
-	"Doña",
-	"M.",
-	"Me",
-	"Pr",
-	"Pr.",
-	"Señor",
-	"Señora"
-]);
-/**
-* Honorifics that are abbreviations: a dot after them is an
-* abbreviation dot (same sentence), so the detector keeps an
-* optional `.` between the title and the name ("Mr. Smith").
-* Every other honorific is a full word — a trailing dot ends a
-* sentence, so the detector must NOT consume it (otherwise a span
-* like "President. The Employee" crosses the sentence boundary).
-*
-* Maintained explicitly, NOT derived from "ends in a dot": "Mr" is
-* an abbreviation written without a dot, and "Lord" is a full word.
-*/
-const HONORIFIC_ABBREVIATION = new Set([
-	"Avv.",
-	"Dott.",
-	"M.",
-	"Me",
-	"Messrs",
-	"Mlle",
-	"Mme",
-	"Mr",
-	"Mrs",
-	"Ms",
-	"Pr",
-	"Pr.",
-	"Sig.",
-	"Sig.ra",
-	"Sr.",
-	"Sra."
-]);
-/**
-* Post-nominal degrees (comma or space separated after name).
-* Plain text; the detector auto-escapes for regex.
-*/
-const POST_NOMINALS = [
-	"Ph.D.",
-	"Ph.D",
-	"CSc.",
-	"DrSc.",
-	"ArtD.",
-	"D.Phil.",
-	"DPhil.",
-	"MPhil.",
-	"MBA",
-	"MPA",
-	"LL.M.",
-	"LL.B.",
-	"M.Sc.",
-	"B.Sc.",
-	"MSc.",
-	"BSc.",
-	"M.Eng.",
-	"B.Eng.",
-	"M.A.",
-	"B.A.",
-	"JCD",
-	"JD",
-	"DiS.",
-	"ACCA",
-	"FCCA",
-	"CIPM",
-	"CIPT",
-	"CIPP/E",
-	"CIPP",
-	"KC",
-	"QC"
-];
-//#endregion
 //#region src/data/amount-words.json
 var amount_words_exports = /* @__PURE__ */ __exportAll({
 	_comment: () => _comment$1,
@@ -10138,6 +10746,8 @@ const DECIMAL_COMMA_RE = new RegExp(`^,(?:\\d|${DASH}{1,2})`);
 * etc.), skip the comma and degree, then continue.
 */
 const POST_NOMINAL_RE = new RegExp(`^,\\s*(?:${POST_NOMINALS.toSorted((a, b) => b.length - a.length).map((d) => d.replace(/[.*+?^${}()|[\]\\]/g, "\\$&").replace(/\\\./g, "\\.\\s*")).join("|")})\\.?`, "i");
+const PERSON_NAME_TOKEN = "\\p{Lu}[\\p{L}\\p{M}.'’\\-]*";
+const PERSON_NAME_RUN_RE = new RegExp(`^${PERSON_NAME_TOKEN}(?:,?[ \\t]+(?:${`(?:${NAME_PARTICLE}[ \\t]+|d['’])`}){0,2}${PERSON_NAME_TOKEN})*`, "u");
 let cachedLegalFormCheckSource = null;
 let cachedLegalFormCheckForms = [];
 const LEGAL_FORM_ALNUM_RE = /[\p{L}\p{N}]/u;
@@ -10530,6 +11140,24 @@ const TRIGGER_LOOKAHEAD_MARGIN = 128;
 const LINE_TRIGGER_LOOKAHEAD = 2048;
 const MATCH_PATTERN_LOOKAHEAD = 512;
 const PHONE_VALUE_START_RE = /^[+(\d]/;
+/**
+* Phone-shaped run: an opening "+", "(", or digit, then
+* any mix of digits, whitespace, parens, dots, slashes,
+* and dash variants. The dash class includes every
+* typographic dash (U+2011 non-breaking hyphen, U+2013
+* en-dash, etc.) via `DASH_INNER` so a Unicode separator
+* ("+1 555‑123‑4567") does not break the run and leave the
+* number unredacted.
+*/
+const PHONE_SHAPE_PREFIX_RE = new RegExp(`^[+(\\d][\\d\\s()./${DASH_INNER}]*`);
+/**
+* Optional phone-extension suffix following the numeric
+* run: "ext", "ext.", "extension", or "x" (case-
+* insensitive) plus 1-6 digits, with optional leading
+* whitespace. Matches "+1 555 123 4567 ext. 89" and
+* "555-1234 x42" so the shape bound keeps the extension.
+*/
+const PHONE_EXTENSION_SUFFIX_RE = /^\s*(?:ext\.?|extension|x)\s*\d{1,6}/i;
 const ISO_DATE_PREFIX_RE = /^\d{4}-\d{2}-\d{2}\b/;
 const INLINE_FIELD_LABEL_RE = /\b[\p{L}][\p{L}\p{M} /-]{1,32}:/u;
 const INLINE_FIELD_LABEL_STOP_RE = /(?:^|[^\S\n\t])[\p{L}][\p{L}\p{M} /-]{1,32}:/u;
@@ -10622,6 +11250,19 @@ const extractValue = (text, triggerEnd, strategy, label) => {
 					end = inlineLabel.index;
 					foundLineStop = true;
 				}
+				const region = valueText.slice(0, end);
+				const sentenceBreak = /\.\s/.exec(region);
+				const shapeRegion = sentenceBreak !== null ? region.slice(0, sentenceBreak.index) : region;
+				const shape = PHONE_SHAPE_PREFIX_RE.exec(shapeRegion);
+				if (shape) {
+					let shapeEnd = shape[0].replace(/\D+$/, "").length;
+					const extension = PHONE_EXTENSION_SUFFIX_RE.exec(region.slice(shapeEnd));
+					if (extension !== null) shapeEnd += extension[0].length;
+					if (shapeEnd > 0 && shapeEnd < end) {
+						end = shapeEnd > MAX_TRIGGER_VALUE_LEN ? Math.min(capAtWordBoundary(valueText, MAX_TRIGGER_VALUE_LEN), MAX_TRIGGER_VALUE_LEN) : shapeEnd;
+						foundLineStop = true;
+					}
+				}
 			}
 			if (!foundLineStop) end = capAtWordBoundary(valueText, Math.min(end, MAX_TRIGGER_VALUE_LEN));
 			const rawSlice = valueText.slice(0, end);
@@ -10800,14 +11441,31 @@ const processTriggerMatches = (allMatches, sliceStart, sliceEnd, fullText, rules
 		if (LETTER_RE.test(triggerLastChar) && LETTER_RE.test(fullText[match.end] ?? "")) continue;
 		const triggerEnd = match.end;
 		const rawValue = extractValue(fullText, triggerEnd, rule.strategy, rule.label);
-		const value = rawValue ? stripQuotes(rawValue) : null;
+		let value = rawValue ? stripQuotes(rawValue) : null;
 		if (value) {
 			if (!applyValidations(value.text, rule.validations)) continue;
 			if (rule.label === "phone number" && !isPlausiblePhoneTriggerValue(value.text)) continue;
+			if (rule.label === "phone number" && value.text.length > MAX_TRIGGER_VALUE_LEN && fullText[value.end] !== "\n" && fullText[value.end] !== "	") {
+				const cappedEnd = Math.min(capAtWordBoundary(value.text, MAX_TRIGGER_VALUE_LEN), MAX_TRIGGER_VALUE_LEN);
+				const cappedText = value.text.slice(0, cappedEnd).trimEnd();
+				if (cappedText.length === 0) continue;
+				value = {
+					...value,
+					end: value.start + cappedText.length,
+					text: cappedText
+				};
+			}
 			const entityStart = rule.includeTrigger ? match.start : value.start;
-			const entityEnd = value.end;
-			const entityText = fullText.slice(entityStart, entityEnd);
+			let entityEnd = value.end;
+			let entityText = fullText.slice(entityStart, entityEnd);
 			const effectiveLabel = rule.label === "person" && hasKnownLegalFormSuffix(entityText) ? "organization" : rule.label;
+			if (effectiveLabel === "person") {
+				const run = PERSON_NAME_RUN_RE.exec(value.text);
+				if (run !== null && run[0].length < value.text.length) {
+					entityEnd = value.start + run[0].length;
+					entityText = fullText.slice(entityStart, entityEnd);
+				}
+			}
 			results.push({
 				start: entityStart,
 				end: entityEnd,
@@ -11431,6 +12089,9 @@ const EMPTY_ALLOW_LIST = /* @__PURE__ */ new Set();
 const getAllowList = (ctx) => ctx.allowList ?? EMPTY_ALLOW_LIST;
 let commonWordsPromise = null;
 let commonWordsCache = null;
+const EMPTY_COMMON_WORDS = /* @__PURE__ */ new Set();
+/** Sync accessor — returns empty set before init. */
+const getCommonWords = () => commonWordsCache ?? EMPTY_COMMON_WORDS;
 const loadCommonWords = () => {
 	if (commonWordsCache) return Promise.resolve(commonWordsCache);
 	if (commonWordsPromise) return commonWordsPromise;
@@ -11691,6 +12352,7 @@ const patternSources = (sources) => {
 	if (sources === void 0) return EMPTY_PATTERN_SOURCES;
 	return Array.isArray(sources) ? sources : [sources];
 };
+const hasPersonNameSource = (match) => match.sources.includes("first-name") || match.sources.includes("surname");
 const addPatternLabel = (list, index, label) => {
 	const existing = list[index];
 	if (existing === void 0) {
@@ -11878,8 +12540,21 @@ const appendNameCorpusEntries = (config, ctx, patternList, labelList, sourceList
 			sourceList.push(source);
 		}
 	};
-	for (const name of getNameCorpusFirstNames(ctx)) addNameEntry(name, "first-name");
-	for (const name of getNameCorpusSurnames(ctx)) addNameEntry(name, "surname");
+	const commonWords = getCommonWords();
+	const addDeclinedVariants = (name, source) => {
+		for (const variant of expandNameDeclensions(name)) {
+			if (commonWords.has(variant.toLowerCase())) continue;
+			addNameEntry(variant, source);
+		}
+	};
+	for (const name of getNameCorpusFirstNames(ctx)) {
+		addNameEntry(name, "first-name");
+		addDeclinedVariants(name, "first-name");
+	}
+	for (const name of getNameCorpusSurnames(ctx)) {
+		addNameEntry(name, "surname");
+		addDeclinedVariants(name, "surname");
+	}
 	for (const title of getNameCorpusTitles(ctx)) {
 		const norm = stripCuratedPatternSyntax(normalizeForSearch(title));
 		if (norm.length === 0) continue;
@@ -12024,6 +12699,7 @@ const processDenyListMatches = (allMatches, sliceStart, sliceEnd, fullText, data
 		const first = chain.at(0);
 		const last = chain.at(-1);
 		if (!first || !last) continue;
+		if (!chain.some(hasPersonNameSource)) continue;
 		if (isSuppressibleDefinedTermQuote(fullText, first.start, ctx)) continue;
 		const extended = extendPersonName(fullText, first.start, last.end, ctx);
 		const score = chain.length >= 2 ? .9 : .5;
@@ -12656,25 +13332,33 @@ const isCallerOwnedEntity$2 = (entity) => entity.sourceDetail === "custom-deny-l
 * to get the base name, and re-scan the full text for
 * bare mentions of that base name. Returns new entities
 * for occurrences not already covered.
+*
+* Propagated mentions are coref aliases: each carries
+* `corefSourceText` linking it to the full seed entity
+* text, so placeholder numbering assigns the bare
+* mention the same placeholder as its source ("Acme"
+* and "Acme Corp." both become [ORGANIZATION_1]).
 */
 const propagateOrgNames = (entities, fullText) => {
-	const seeds = [];
-	const seenBases = /* @__PURE__ */ new Set();
+	const seedByBase = /* @__PURE__ */ new Map();
 	for (const e of entities) {
 		if (e.label !== "organization") continue;
 		if (isCallerOwnedEntity$2(e)) continue;
 		for (const suffix of LEGAL_SUFFIXES) if (e.text.endsWith(suffix)) {
 			const base = e.text.slice(0, -suffix.length).replace(TRAILING_SEP, "").trim();
-			if (base.length >= 3 && !seenBases.has(base)) {
-				seenBases.add(base);
-				seeds.push({
+			if (base.length >= 3) {
+				const existing = seedByBase.get(base);
+				if (existing === void 0) seedByBase.set(base, {
 					baseName: base,
-					label: e.label
+					label: e.label,
+					sourceText: e.text
 				});
+				else if (existing.sourceText !== e.text) existing.sourceText = base;
 			}
 			break;
 		}
 	}
+	const seeds = [...seedByBase.values()];
 	if (seeds.length === 0) return [];
 	const covered = entities.map((e) => [e.start, e.end]);
 	const isOverlapping = (start, end) => covered.some(([cs, ce]) => start < ce && end > cs);
@@ -12708,7 +13392,8 @@ const propagateOrgNames = (entities, fullText) => {
 					label,
 					text: fullText.slice(spanStart, matchEnd),
 					score: ORG_PROPAGATION_SCORE,
-					source: DETECTION_SOURCES.COREFERENCE
+					source: DETECTION_SOURCES.COREFERENCE,
+					corefSourceText: seed.sourceText
 				});
 				covered.push([spanStart, matchEnd]);
 			}
@@ -13597,6 +14282,7 @@ const applyHotwordRules = (entities, fullText) => {
 /** Max gap (in chars) between entities to merge. */
 const MAX_GAP = 3;
 const hasLockedBoundary$1 = (entity) => entity.sourceDetail === "custom-deny-list" || entity.sourceDetail === "custom-regex";
+const hasDetectorLockedBoundary = (entity) => entity.label === "phone number" && entity.source === DETECTION_SOURCES.TRIGGER;
 /**
 * Characters allowed in the gap between two adjacent
 * same-label entities that should be merged: spaces,
@@ -13777,7 +14463,7 @@ const fixPartialWords = (entities, fullText) => {
 	})).sort((a, b) => a.entity.end - b.entity.end);
 	const endPositions = byEnd.map((x) => x.entity.end);
 	return sorted.map((e, eIdx) => {
-		if (hasLockedBoundary$1(e)) return e;
+		if (hasLockedBoundary$1(e) || hasDetectorLockedBoundary(e)) return e;
 		if (e.text !== fullText.slice(e.start, e.end)) return e;
 		let newStart = wordStartAt(e.start, boundaries, fullText);
 		let newEnd = wordEndAt(e.end, boundaries, fullText);
@@ -14180,6 +14866,8 @@ const unmaskNerEntities = (nerEntities, maskResult, fullText) => {
 const LITERAL_SOURCES = new Set(["deny-list", "gazetteer"]);
 const isCallerOwnedEntity = (entity) => entity.sourceDetail === "custom-deny-list" || entity.sourceDetail === "custom-regex";
 const hasLockedBoundary = (entity) => isCallerOwnedEntity(entity);
+const isCoveredByDenyListEntity = (entity, denyListEntity) => entity.label === denyListEntity.label && denyListEntity.start <= entity.start && denyListEntity.end >= entity.end;
+const filterNameCorpusCoveredByDenyList = (nameCorpusEntities, denyListEntities) => nameCorpusEntities.filter((entity) => !denyListEntities.some((denyListEntity) => isCoveredByDenyListEntity(entity, denyListEntity)));
 const LITERAL_BOUNDARY_PUNCT_RE = /^["“„‟‘‛'«]|["”’'»!.]$/u;
 const BARE_POSTAL_CODE_RE = /^\s*(?:\d{3}\s?\d{2}|\d{2}[-–]\d{3}|\d{5}(?:[-–]\d{3,4})?)\s*$/u;
 const hasCuratedLiteralBoundary = (entity) => LITERAL_SOURCES.has(entity.source) && entity.label !== "person" && entity.sourceDetail !== "gazetteer-extension" && LITERAL_BOUNDARY_PUNCT_RE.test(entity.text);
@@ -14700,6 +15388,13 @@ const runPipeline = async (options) => {
 	const rawDenyListEntities = config.enableDenyList && search.denyListData ? processDenyListMatches(literalMatches, slices.denyList.start, slices.denyList.end, fullText, search.denyListData, ctx) : [];
 	const denyListEntities = filterAllowedLabels(rawDenyListEntities, preHotwordAllowedLabels);
 	if (denyListEntities.length > 0) log("deny-list", `${denyListEntities.length} matches`);
+	if (config.enableNameCorpus && config.enableDenyList) {
+		await initNameCorpus(ctx, config.dictionaries, config.nameCorpusLanguages);
+		checkAbort(signal);
+		rawNameCorpusEntities = filterNameCorpusCoveredByDenyList(detectNameCorpus(fullText, ctx, { mode: "supplemental" }), rawDenyListEntities);
+		nameCorpusEntities = filterAllowedLabels(rawNameCorpusEntities, preHotwordAllowedLabels);
+		log("name-corpus", `${nameCorpusEntities.length} supplemental matches`);
+	}
 	const rawGazetteerEntities = config.enableGazetteer && search.gazetteerData ? processGazetteerMatches(literalMatches, slices.gazetteer.start, slices.gazetteer.end, fullText, search.gazetteerData) : [];
 	const gazetteerEntities = filterAllowedLabels(rawGazetteerEntities, preHotwordAllowedLabels);
 	if (gazetteerEntities.length > 0) log("gazetteer", `${gazetteerEntities.length} matches`);
@@ -14797,7 +15492,6 @@ const runPipeline = async (options) => {
 	const merged = filterFalsePositives(postOrgEntities, ctx, fullText);
 	if (merged.length < postOrgEntities.length) log("filter", `removed ${postOrgEntities.length - merged.length} FPs`);
 	checkAbort(signal);
-	ctx.corefSourceMap.clear();
 	if (config.enableCoreference) {
 		if (!legalFormsEnabled) await warmLegalRoleHeads();
 		const terms = await extractDefinedTerms(fullText, merged.filter((entity) => !isCallerOwnedEntity(entity)), ctx);
@@ -14863,12 +15557,11 @@ const normalizeEntityText = (label, text) => {
 * Placeholder format: [LABEL_N] where LABEL is uppercase
 * and N is a 1-based counter per label.
 *
-* @param ctx Pipeline context. Must be the same instance
-*   passed to `runPipeline` (or `findCoreferenceSpans`)
-*   so coreference placeholder links are preserved.
-*   Defaults to `defaultContext` for single-tenant usage.
+* @param _ctx Unused. Kept for signature compatibility;
+*   coref alias links now travel on the entities
+*   themselves (`corefSourceText`).
 */
-const buildPlaceholderMap = (entities, ctx = defaultContext) => {
+const buildPlaceholderMap = (entities, _ctx = defaultContext) => {
 	const counters = /* @__PURE__ */ new Map();
 	const textLabelToPlaceholder = /* @__PURE__ */ new Map();
 	const normalizedToPlaceholder = /* @__PURE__ */ new Map();
@@ -14877,9 +15570,9 @@ const buildPlaceholderMap = (entities, ctx = defaultContext) => {
 		const compositeKey = `${entity.label}\0${entity.text}`;
 		if (textLabelToPlaceholder.has(compositeKey)) continue;
 		const labelKey = entity.label.toUpperCase().replace(WHITESPACE_RE, "_");
-		const sourceText = ctx.corefSourceMap.get(corefKey(entity));
-		if (sourceText !== void 0) {
-			const sourceNormalizedKey = `${labelKey}\0${normalizeEntityText(entity.label, sourceText)}`;
+		const sourceText = entity.source === "coreference" ? entity.corefSourceText : void 0;
+		const sourceNormalizedKey = sourceText === void 0 ? void 0 : `${labelKey}\0${normalizeEntityText(entity.label, sourceText)}`;
+		if (sourceNormalizedKey !== void 0) {
 			const sourceExisting = normalizedToPlaceholder.get(sourceNormalizedKey);
 			if (sourceExisting) {
 				textLabelToPlaceholder.set(compositeKey, sourceExisting);
@@ -14890,6 +15583,7 @@ const buildPlaceholderMap = (entities, ctx = defaultContext) => {
 		const existing = normalizedToPlaceholder.get(normalizedKey);
 		if (existing) {
 			textLabelToPlaceholder.set(compositeKey, existing);
+			if (sourceNormalizedKey !== void 0) normalizedToPlaceholder.set(sourceNormalizedKey, existing);
 			continue;
 		}
 		const count = (counters.get(labelKey) ?? 0) + 1;
@@ -14897,6 +15591,7 @@ const buildPlaceholderMap = (entities, ctx = defaultContext) => {
 		const placeholder = `[${labelKey}_${count}]`;
 		textLabelToPlaceholder.set(compositeKey, placeholder);
 		normalizedToPlaceholder.set(normalizedKey, placeholder);
+		if (sourceNormalizedKey !== void 0) normalizedToPlaceholder.set(sourceNormalizedKey, placeholder);
 	}
 	return textLabelToPlaceholder;
 };
@@ -14939,7 +15634,7 @@ const redactText = (fullText, entities, config = DEFAULT_OPERATOR_CONFIG, ctx =
 		const replacement = operator.apply(entity.text, entity.label, placeholder, config.redactString);
 		parts.push(replacement);
 		operatorMap.set(placeholder, opType);
-		if (operator.reversibility === "reversible" && !redactionMap.has(placeholder)) redactionMap.set(placeholder, entity.text);
+		if (operator.reversibility === "reversible" && !redactionMap.has(placeholder)) redactionMap.set(placeholder, entity.source === "coreference" ? entity.corefSourceText : entity.text);
 		cursor = entity.end;
 	}
 	if (cursor < fullText.length) parts.push(fullText.slice(cursor));
@@ -15395,6 +16090,6 @@ const levenshtein = (rawA, rawB) => {
 //#region src/wasm.ts
 initTextSearch(TextSearch);
 //#endregion
-export { CURRENCY_PATTERN_META, DATE_PATTERN_META, DEFAULT_ENTITY_LABELS, DEFAULT_OPERATOR_CONFIG, DETECTION_SOURCES, DETECTOR_PRIORITY, OPERATOR_REGISTRY, OPERATOR_TYPES, REGEX_META, REGEX_PATTERNS, REGIONS, ZONE_SCORE_ADJUSTMENTS, applyHotwordRules, applyZoneAdjustments, boostNearMissEntities, buildDenyList, buildGazetteerPatterns, buildLegalFormPatterns, buildPlaceholderMap, buildStreetTypePatterns, buildTriggerPatterns, buildUnifiedSearch, chunkText, classifyZones, computeChunkOffsets, corefKey, createPipelineContext, deanonymise, decodeSpans, decodeTokenSpans, detectNameCorpus, ensureDenyListData, exportRedactionKey, extractDefinedTerms, filterFalsePositives, findCoreferenceSpans, getCurrencyPatterns, getDatePatterns, initAddressComponents, initHotwordRules, initNameCorpus, initZoneClassifier, levenshtein, mergeAndDedup, mergeChunkEntities, normalizeForSearch, prepareBatch, preparePipelineSearch, processAddressSeeds, processDenyListMatches, processGazetteerMatches, processLegalFormMatches, processRegexMatches, processTriggerMatches, propagateOrgNames, redactText, resolveCountries, resolveOperator, runPipeline, runUnifiedSearch, sanitizeEntities, tokenizeText, warmLegalRoleHeads };
+export { CURRENCY_PATTERN_META, DATE_PATTERN_META, DEFAULT_ENTITY_LABELS, DEFAULT_OPERATOR_CONFIG, DETECTION_SOURCES, DETECTOR_PRIORITY, OPERATOR_REGISTRY, OPERATOR_TYPES, REGEX_META, REGEX_PATTERNS, REGIONS, ZONE_SCORE_ADJUSTMENTS, applyHotwordRules, applyZoneAdjustments, boostNearMissEntities, buildDenyList, buildGazetteerPatterns, buildLegalFormPatterns, buildPlaceholderMap, buildStreetTypePatterns, buildTriggerPatterns, buildUnifiedSearch, chunkText, classifyZones, computeChunkOffsets, corefKey, createPipelineContext, deanonymise, decodeSpans, decodeTokenSpans, detectNameCorpus, ensureDenyListData, exportRedactionKey, extractDefinedTerms, filterFalsePositives, findCoreferenceSpans, getCurrencyPatterns, getDatePatterns, getNameCorpusNonWesternNames, initAddressComponents, initHotwordRules, initNameCorpus, initZoneClassifier, levenshtein, mergeAndDedup, mergeChunkEntities, normalizeForSearch, prepareBatch, preparePipelineSearch, processAddressSeeds, processDenyListMatches, processGazetteerMatches, processLegalFormMatches, processRegexMatches, processTriggerMatches, propagateOrgNames, redactText, resolveCountries, resolveOperator, runPipeline, runUnifiedSearch, sanitizeEntities, tokenizeText, warmLegalRoleHeads };
 
 //# sourceMappingURL=wasm.mjs.map