@stll/anonymize 1.0.3 → 1.0.5

package/dist/index.mjs

@@ -73,6 +73,7 @@ const DEFAULT_ENTITY_LABELS = [
 	"monetary amount",
 	"land parcel"
 ];
+const isLegalFormsEnabled = (config) => config.enableLegalForms !== false;
 //#endregion
 //#region src/context.ts
 /**
@@ -293,6 +294,33 @@ const getDefinitionPatterns = async (ctx) => {
 };
 const SEARCH_WINDOW = 200;
 /**
+* Check whether an alias has textual similarity to
+* the source entity. Prevents roles and structural
+* terms from being treated as name aliases.
+*
+* Three checks (any passes → similar):
+* 1. Word overlap: a word in the alias appears in the
+*    entity (case-insensitive, min 2 chars)
+* 2. Initials: alias letters match first letters of
+*    entity words ("TB" ↔ "Tomas Bata")
+* 3. Substring: alias is a substring of the entity
+*    or vice versa (min 3 chars)
+*/
+const hasEntitySimilarity = (alias, entityText) => {
+	const aliasLower = alias.toLowerCase();
+	const entityLower = entityText.toLowerCase();
+	if (aliasLower.length >= 3 && entityLower.includes(aliasLower)) return true;
+	if (entityLower.length >= 3 && aliasLower.includes(entityLower)) return true;
+	const aliasWords = aliasLower.split(/[\s.,;:'"()/-]+/).filter((w) => w.length >= 2);
+	const entityWords = entityLower.split(/[\s.,;:'"()/-]+/).filter((w) => w.length >= 2);
+	const entityWordSet = new Set(entityWords);
+	for (const word of aliasWords) if (entityWordSet.has(word)) return true;
+	if (/^[\p{Lu}]+$/u.test(alias) && alias.length >= 2 && alias.length <= entityWords.length) {
+		for (let start = 0; start <= entityWords.length - alias.length; start++) if (entityWords.slice(start, start + alias.length).map((w) => w.charAt(0)).join("") === aliasLower) return true;
+	}
+	return false;
+};
+/**
 * Scan for defined-term patterns near known entities.
 *
 * Legal documents universally follow:
@@ -332,6 +360,9 @@ const extractDefinedTerms = async (fullText, entities, ctx = defaultContext) =>
 				break;
 			}
 			if (bestEntity === null) continue;
+			const gapText = fullText.slice(bestEntity.end, defPos);
+			if (/(?:;|\.(?=\s*(?:["'„‚(]*\p{Lu}|$)))/u.test(gapText)) continue;
+			if (!hasEntitySimilarity(alias, bestEntity.text)) continue;
 			const key = `${alias.toLowerCase()}::${bestEntity.label}`;
 			if (seen.has(key)) continue;
 			seen.add(key);
@@ -697,6 +728,8 @@ const TOKEN_TYPE = {
 	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.
@@ -828,7 +861,9 @@ const detectNameCorpus = (fullText, ctx = defaultContext) => {
 			if (!next) break;
 			const prev = chain.at(-1);
 			if (prev) {
-				if (fullText.slice(prev.end, next.start).includes("\n")) break;
+				const gap = fullText.slice(prev.end, next.start);
+				const breaksOnPeriod = gap.includes(".") && !isInitialContinuationGap$1(prev.text, gap);
+				if (gap.includes("\n") || PERSON_CHAIN_BREAK_RE$1.test(gap) || breaksOnPeriod) break;
 			}
 			if (next.type === TOKEN_TYPE.NAME || next.type === TOKEN_TYPE.SURNAME || next.type === TOKEN_TYPE.TITLE || next.type === TOKEN_TYPE.ABBREVIATION || next.type === TOKEN_TYPE.CAPITALIZED) {
 				chain.push(next);
@@ -1396,6 +1431,7 @@ const buildDatePatternsFromMonths = (alt) => {
 	return [
 		`(?i)\\b\\d{1,2}\\.?\\s+(?:${alt})\\.?\\s+\\d{4}(?:\\s+\\d{1,2}:\\d{2}(?::\\d{2})?)?\\b`,
 		`(?i)\\b(?:${alt})\\.?\\s+\\d{1,2},?\\s+\\d{4}\\b`,
+		`(?i)\\b(?:${alt})\\.?\\s+\\d{1,2}(?=\\s|[.,;!?)]|$)`,
 		`(?i)\\b\\d{1,2}(?:st|nd|rd|th)\\s+(?:${alt})\\.?(?:\\s+\\d{4})?(?=\\s|[.,;!?)]|$)`,
 		`(?i)\\b(?:${alt})\\.?\\s+\\d{4}\\b`,
 		`(?i)\\b\\d{4}\\.\\s+(?:${alt})\\.?\\s+\\d{1,2}\\.?(?=\\s|[.,;!?)]|$)`,
@@ -1462,9 +1498,11 @@ const buildCurrencyPatterns = (data) => {
 	if (!symbols && !trailingAlt) return [];
 	const NUM = "(?:\\d{1,3}(?:[,.'[^\\S\\n\\t]]\\d{3})+|\\d{1,9})";
 	const patterns = [];
-	if (symbols) patterns.push(`(?:[${symbols}])[^\\S\\n\\t]?${NUM}(?:[.,][^\\S\\n\\t]?(?:\\d{1,2}${DASH}?|${DASH}{1,2})?)?\\b`);
-	if (trailingAlt) patterns.push(`\\b(?:${trailingAlt})[^\\S\\n\\t]{0,2}${NUM}(?:[.,][^\\S\\n\\t]?(?:\\d{1,2}${DASH}?|${DASH}{1,2})?)?\\b`);
-	if (trailingAlt) patterns.push(`\\b${NUM}(?:[.,][^\\S\\n\\t]?(?:\\d{1,2}${DASH}?|${DASH}{1,2})?)?[^\\S\\n\\t]{0,4}(?:${trailingAlt})(?:\\b|(?=\\s|[.,;!?)]|$))`);
+	const DECIMAL = `(?:[.,](?=\\d|[${DASH_INNER}])[^\\S\\n\\t]?(?:\\d{1,2}${DASH}?|${DASH}{1,2}))?`;
+	const END = `(?:\\b|(?=\\s|[.,;!?)]|$))`;
+	if (symbols) patterns.push(`(?:[${symbols}])[^\\S\\n\\t]?${NUM}${DECIMAL}${END}`);
+	if (trailingAlt) patterns.push(`\\b(?:${trailingAlt})[^\\S\\n\\t]{0,2}${NUM}${DECIMAL}${END}`);
+	if (trailingAlt) patterns.push(`\\b${NUM}${DECIMAL}[^\\S\\n\\t]{0,4}(?:${trailingAlt})${END}`);
 	return patterns;
 };
 /** Cached promise for currency patterns. Loaded once. */
@@ -1572,10 +1610,10 @@ const ALLCAP_WORD = `[${UPPER}]{2,}`;
 const ROMAN_NUMERAL_RE = /^(?=[IVXLCDM])M{0,3}(?:CM|CD|D?C{0,3})(?:XC|XL|L?X{0,3})(?:IX|IV|V?I{0,3})$/;
 const escapeForRegex = (form) => form.replace(/[.*+?^${}()|[\]\\]/g, "\\$&").replace(/\s+/g, "\\s+").replace(/\\\./g, "\\.[^\\S\\n]?");
 const isShortForm = (form) => form.replace(/[.\s]/g, "").length <= 3 && !form.includes(" ");
-const buildPatternString = (forms, requireCapBefore) => {
+const buildPatternString = (forms) => {
 	if (forms.length === 0) return null;
 	const alt = forms.toSorted((a, b) => b.length - a.length).map(escapeForRegex).join("|");
-	return `${`(?:${CAP_WORD})(?:${`(?:[\\s&,.${DASH_INNER}]{1,4}|\\s+(?:a|and|und|et|e|y|i)\\s+)`}(?:${ANY_WORD})){0,10}`}${requireCapBefore ? `(?:\\s+|,\\s*)` : `\\s+`}(?:${alt})(?![${LOWER}])`;
+	return `${`(?:${CAP_WORD})(?:${`(?:[\\s&,.${DASH_INNER}]{1,4}|${`\\s+(?:a|and|und|et|e|y|i)\\s+(?=[${LOWER}])`})`}(?:${ANY_WORD})){0,10}`}(?:\\s+|,\\s*)(?:${alt})(?![${LOWER}])`;
 };
 /**
 * Build legal form regex pattern strings.
@@ -1600,15 +1638,82 @@ const buildLegalFormPatterns = async () => {
 		}
 	}
 	const patterns = [];
-	const longPattern = buildPatternString(allForms.filter((f) => !isShortForm(f)), false);
+	const longPattern = buildPatternString(allForms.filter((f) => !isShortForm(f)));
 	if (longPattern) patterns.push(longPattern);
-	const shortPattern = buildPatternString(allForms.filter(isShortForm), true);
+	const shortPattern = buildPatternString(allForms.filter(isShortForm));
 	if (shortPattern) patterns.push(shortPattern);
 	const allcapPrefix = `(?:${ALLCAP_WORD})(?:[\\s&,.${DASH_INNER}]{1,4}(?:${ALLCAP_WORD})){0,2}`;
 	const allcapAlt = allForms.toSorted((a, b) => b.length - a.length).map(escapeForRegex).join("|");
 	patterns.push(`${allcapPrefix}(?:\\s+|,\\s*)(?:${allcapAlt})(?![${LOWER}])`);
 	return patterns;
 };
+const CONNECTOR_RE = /^(?:a|and|und|et|e|y|i|&)$/i;
+const LEADING_CLAUSE_RE = /(?:^|\s)(?:by\s+and\s+between|is\s+between)\s+/giu;
+/**
+* Find the word ending just before `pos` in `text`,
+* skipping any whitespace (not newlines).
+* Returns null if no word is found (e.g., at start
+* of text, or preceded by non-word chars like ".").
+*/
+const findWordBefore = (text, pos) => {
+	let scan = pos - 1;
+	while (scan >= 0) {
+		const ch = text.charAt(scan);
+		if (ch === "\n" || !/\s/.test(ch)) break;
+		scan--;
+	}
+	if (scan < 0 || text.charAt(scan) === "\n") return null;
+	const wordEnd = scan + 1;
+	while (scan >= 0 && /[\p{L}\p{M}&]/u.test(text.charAt(scan))) scan--;
+	const wordStart = scan + 1;
+	const word = text.slice(wordStart, wordEnd);
+	if (word.length === 0) return null;
+	return {
+		word,
+		start: wordStart
+	};
+};
+/**
+* Extend a match backward through uppercase words and
+* lowercase connectors. Stops at start of text,
+* newline, or a word that doesn't qualify.
+*
+* Connectors (a, and, und, et, ...) are only consumed
+* when there is a valid word before them — a trailing
+* connector at an entity boundary is not consumed.
+*/
+const extendBackward = (fullText, matchStart) => {
+	let pos = matchStart;
+	while (pos > 0) {
+		const found = findWordBefore(fullText, pos);
+		if (!found) break;
+		const { word, start: wordStart } = found;
+		const isUpper = /^\p{Lu}/u.test(word);
+		const isConnector = CONNECTOR_RE.test(word);
+		if (isUpper) pos = wordStart;
+		else if (isConnector) {
+			const prev = findWordBefore(fullText, wordStart);
+			if (!prev) break;
+			if (!/^\p{Lu}/u.test(prev.word)) break;
+			pos = prev.start;
+		} else break;
+	}
+	return pos;
+};
+const trimLeadingClause = (text) => {
+	let cut = -1;
+	for (const match of text.matchAll(LEADING_CLAUSE_RE)) cut = match.index + match[0].length;
+	if (cut <= 0) return {
+		offset: 0,
+		text
+	};
+	const trimmed = text.slice(cut);
+	const leadingWs = trimmed.match(/^\s*/u)?.[0].length ?? 0;
+	return {
+		offset: cut + leadingWs,
+		text: trimmed.slice(leadingWs)
+	};
+};
 /**
 * Process legal form matches from the unified search.
 * Receives all matches; filters to the legal forms
@@ -1622,27 +1727,52 @@ const processLegalFormMatches = (allMatches, sliceStart, sliceEnd, fullText) =>
 		const text = match.text.trimEnd();
 		if (text.length < 5) continue;
 		if (text.includes("\n")) continue;
-		const prefixEnd = text.lastIndexOf(",") !== -1 ? text.lastIndexOf(",") : text.lastIndexOf(" ");
-		const prefixPart = prefixEnd > 0 ? text.slice(0, prefixEnd).replace(/[^a-zA-ZÀ-ž]/g, "") : text.replace(/[^a-zA-ZÀ-ž]/g, "");
-		const isAllCapsMatch = prefixPart.length > 2 && prefixPart === prefixPart.toUpperCase();
+		let entityStart = match.start;
+		let entityText = text;
+		if (fullText) {
+			const extended = extendBackward(fullText, match.start);
+			if (extended < match.start) {
+				entityStart = extended;
+				entityText = fullText.slice(extended, match.start + text.length).trimEnd();
+			}
+		}
+		const clauseTrim = trimLeadingClause(entityText);
+		if (clauseTrim.offset > 0) {
+			entityStart += clauseTrim.offset;
+			entityText = clauseTrim.text;
+		}
+		const getPrefixInfo = (value) => {
+			const prefixEnd = value.lastIndexOf(",") !== -1 ? value.lastIndexOf(",") : value.lastIndexOf(" ");
+			return {
+				prefixEnd,
+				prefixPart: prefixEnd > 0 ? value.slice(0, prefixEnd).replace(/[^a-zA-ZÀ-ž]/g, "") : value.replace(/[^a-zA-ZÀ-ž]/g, "")
+			};
+		};
+		let { prefixEnd, prefixPart } = getPrefixInfo(entityText);
+		let isAllCapsMatch = prefixPart.length > 2 && prefixPart === prefixPart.toUpperCase();
 		if (isAllCapsMatch && fullText) {
-			const lineStart = fullText.lastIndexOf("\n", match.start);
-			const lineEnd = fullText.indexOf("\n", match.end);
+			const lineStart = fullText.lastIndexOf("\n", entityStart);
+			const lineEnd = fullText.indexOf("\n", entityStart + entityText.length);
 			const lineLetters = fullText.slice(lineStart + 1, lineEnd === -1 ? fullText.length : lineEnd).replace(/[^a-zA-ZÀ-ž]/g, "");
 			const upperCount = [...lineLetters].filter((c) => c === c.toUpperCase()).length;
 			if (lineLetters.length > 5 && upperCount / lineLetters.length >= .95) continue;
-			if ((prefixPart.length > 0 ? text.slice(0, prefixEnd > 0 ? prefixEnd : text.length).trim().split(/\s+/).length : 0) > 3) continue;
+			if ((prefixPart.length > 0 ? entityText.slice(0, prefixEnd > 0 ? prefixEnd : entityText.length).trim().split(/\s+/).length : 0) > 3) {
+				entityStart = match.start;
+				entityText = text;
+				({prefixEnd, prefixPart} = getPrefixInfo(entityText));
+				isAllCapsMatch = prefixPart.length > 2 && prefixPart === prefixPart.toUpperCase();
+			}
 		} else if (isAllCapsMatch) continue;
-		const lastSpace = text.lastIndexOf(" ");
-		const rawSuffix = lastSpace !== -1 ? text.slice(lastSpace + 1) : "";
+		const lastSpace = entityText.lastIndexOf(" ");
+		const rawSuffix = lastSpace !== -1 ? entityText.slice(lastSpace + 1) : "";
 		const suffixClean = rawSuffix.replace(/[.,]/g, "");
 		if (suffixClean.length > 0 && ROMAN_NUMERAL_RE.test(suffixClean)) continue;
-		if (suffixClean.length <= 2 && !/\./.test(rawSuffix) && /[^\x00-\x7F]/.test(text.slice(0, lastSpace !== -1 ? lastSpace : text.length))) continue;
+		if (suffixClean.length <= 2 && !/\./.test(rawSuffix) && /[^\x00-\x7F]/.test(entityText.slice(0, lastSpace !== -1 ? lastSpace : entityText.length))) continue;
 		results.push({
-			start: match.start,
-			end: match.start + text.length,
+			start: entityStart,
+			end: entityStart + entityText.length,
 			label: "organization",
-			text,
+			text: entityText,
 			score: .95,
 			source: DETECTION_SOURCES.LEGAL_FORM
 		});
@@ -1818,6 +1948,33 @@ const buildTriggerPatterns = async () => {
 	} catch (err) {
 		if (!(err instanceof Error) || !err.message.includes("Cannot find module")) throw err;
 	}
+	try {
+		const yearMod = await import("@stll/anonymize-data/config/year-words.json");
+		const data = yearMod.default ?? yearMod;
+		const seen = /* @__PURE__ */ new Set();
+		const yearValidation = compileValidations([{
+			type: "matches-pattern",
+			pattern: "^(?:19|20)\\d{2}\\.?$"
+		}]);
+		for (const [key, words] of Object.entries(data)) {
+			if (key.startsWith("_") || !Array.isArray(words)) continue;
+			for (const word of words) {
+				const lc = word.toLowerCase();
+				if (seen.has(lc)) continue;
+				seen.add(lc);
+				rules.push({
+					trigger: word,
+					label: "date",
+					strategy: {
+						type: "n-words",
+						count: 1
+					},
+					validations: yearValidation,
+					includeTrigger: false
+				});
+			}
+		}
+	} catch {}
 	const seen = /* @__PURE__ */ new Map();
 	for (const rule of rules) {
 		const key = rule.trigger.toLowerCase();
@@ -1851,7 +2008,11 @@ const stripQuotes = (value) => {
 	};
 };
 /** Hard stop characters for to-next-comma scanning. */
-const COMMA_STOP_CHARS = new Set(["\n", "("]);
+const COMMA_STOP_CHARS = new Set([
+	"\n",
+	"(",
+	"	"
+]);
 /**
 * Field-label keywords that terminate address scanning.
 * When a comma in the address strategy is followed by
@@ -2344,6 +2505,56 @@ const MAX_ENTITY_LENGTH = {
 };
 const SECTION_NUMBER_RE = /^(?:§\s*)?\d{1,3}(?:\.\d{1,3}){0,4}\.?$/;
 const STANDALONE_YEAR_RE = /^(?:19|20)\d{2}$/;
+const NUMBER_ABBREV_RE = /(?:^|[\s(])(?:č|čís|nr|no|n)\.\s*$/i;
+const SIGNING_CLAUSE_ADDRESS_RE = /^(?:v|ve)\s+[^\d,\n]{1,40},?\s+dne$/iu;
+const PERSON_TRAILING_NOUNS = new Set([
+	"association",
+	"period",
+	"reform"
+]);
+const LEGAL_FORM_HEADING_RE = /\b(?:agreement|amendment|contract|exhibit)\b/iu;
+const LEADING_ARTIFACT_RE = /^(?:\.\s)+/u;
+const ADDRESS_ROLE_PREFIX_RE = /^(?:prodávajícího|kupujícího|objednatele|zhotovitele|pronajímatele|dodavatele|odběratele|zaměstnance|zaměstnavatele|nájemce)\s+(?=(?:\p{Lu}|\d|ul\.?|ulice|nám\.?|náměstí|tř\.?|třída|nábř\.?|nábřeží|č\.p\.?|č\.ev\.?|sídliště))/iu;
+const ADDRESS_INLINE_ABBREV_AFTER_RE = /^(?:\p{Lu}[\p{L}\p{M}]{0,3}\.|ul\.?|nám\.?|tř\.?|nábř\.?|č\.p\.?|č\.ev\.?)/u;
+const ADDRESS_INLINE_ABBREV_BEFORE_RE = /(?:^|[\s,])(?:st|ave|rd|dr|blvd|ln|hwy|pkwy|cir|ct|pl|sq|ter|trl|ste|apt|bldg|fl|ul|nám|tř|nábř|č\.p|č\.ev)$/iu;
+const ADDRESS_CONTINUATION_WORD_RE = /^(?:suite|building|floor|unit|apartment|room|tower|wing|block|bldg|ste|apt|fl)\b/iu;
+const trimTrailingAddressProse = (text) => {
+	for (const match of text.matchAll(/\.(?=\s+\p{Lu})/gu)) {
+		const cutoff = match.index;
+		if (cutoff === void 0) continue;
+		const before = text.slice(0, cutoff);
+		if (!HAS_DIGIT_RE.test(before)) continue;
+		const after = text.slice(cutoff + 1).trimStart();
+		if (after.length < 5 || ADDRESS_INLINE_ABBREV_AFTER_RE.test(after) || ADDRESS_INLINE_ABBREV_BEFORE_RE.test(before.trimEnd()) || ADDRESS_CONTINUATION_WORD_RE.test(after)) continue;
+		return before.trimEnd();
+	}
+	return text;
+};
+const normalizeEntity = (entity) => {
+	let start = entity.start;
+	let text = entity.text;
+	const trimLeading = (re) => {
+		const match = re.exec(text);
+		if (!match) return;
+		start += match[0].length;
+		text = text.slice(match[0].length);
+	};
+	trimLeading(LEADING_ARTIFACT_RE);
+	trimLeading(/^\s+/u);
+	if (entity.label === "address") {
+		trimLeading(ADDRESS_ROLE_PREFIX_RE);
+		text = trimTrailingAddressProse(text);
+	}
+	const trailingMatch = /[,\s]+$/u.exec(text);
+	if (trailingMatch) text = text.slice(0, text.length - trailingMatch[0].length);
+	if (text.length === 0) return null;
+	return {
+		...entity,
+		start,
+		end: start + text.length,
+		text
+	};
+};
 const EMPTY_GENERIC_ROLES = /* @__PURE__ */ new Set();
 /**
 * Load generic-roles.json and cache the result on the
@@ -2377,21 +2588,32 @@ const getGenericRoles = (ctx) => ctx.genericRoles ?? EMPTY_GENERIC_ROLES;
 * Runs as a post-processing step after all detection
 * layers have merged.
 */
-const filterFalsePositives = (entities, ctx = defaultContext) => {
+const filterFalsePositives = (entities, ctx = defaultContext, fullText) => {
 	const filtered = [];
 	const roles = getGenericRoles(ctx);
 	for (const entity of entities) {
-		const trimmed = entity.text.trim();
+		const normalized = normalizeEntity(entity);
+		if (!normalized) continue;
+		const trimmed = normalized.text;
 		if (TEMPLATE_PLACEHOLDER_RE.test(trimmed)) continue;
-		const maxLen = MAX_ENTITY_LENGTH[entity.label];
-		if (maxLen && trimmed.length > maxLen && entity.source !== "legal-form") continue;
-		if (SECTION_NUMBER_RE.test(trimmed) && entity.source !== "trigger") continue;
-		if (STANDALONE_YEAR_RE.test(trimmed)) continue;
-		if (entity.label === "person" && HAS_DIGIT_RE.test(trimmed)) continue;
-		if ((entity.label === "person" || entity.label === "organization") && roles.has(trimmed.toLowerCase())) continue;
-		if (entity.label === "address" && trimmed.length > 40 && !POSTAL_CODE_RE.test(trimmed) && !HAS_DIGIT_RE.test(trimmed) && !ADDRESS_COMPONENTS_RE.test(trimmed) && !JURISDICTION_RE.test(trimmed)) continue;
-		if (entity.label === "address" && entity.source === "trigger" && !HAS_DIGIT_RE.test(trimmed) && !ADDRESS_COMPONENTS_RE.test(trimmed) && !JURISDICTION_RE.test(trimmed)) continue;
-		filtered.push(entity);
+		const maxLen = MAX_ENTITY_LENGTH[normalized.label];
+		if (maxLen && trimmed.length > maxLen && normalized.source !== "legal-form") continue;
+		if (SECTION_NUMBER_RE.test(trimmed) && normalized.source !== "trigger") continue;
+		if (STANDALONE_YEAR_RE.test(trimmed) && normalized.source !== "trigger") continue;
+		if (fullText && normalized.source !== "trigger" && /^\d/.test(trimmed) && NUMBER_ABBREV_RE.test(fullText.slice(Math.max(0, normalized.start - 10), normalized.start))) continue;
+		if (normalized.label === "registration number" && /^[\p{L}]{1,2}$/u.test(trimmed)) continue;
+		if (normalized.label === "person" && HAS_DIGIT_RE.test(trimmed)) continue;
+		if (normalized.label === "person") {
+			const tokens = trimmed.split(/\s+/u);
+			const last = tokens.at(-1)?.replace(/[.,;:!?]+$/u, "").toLowerCase();
+			if (tokens.length > 1 && last && PERSON_TRAILING_NOUNS.has(last)) continue;
+		}
+		if ((normalized.label === "person" || normalized.label === "organization") && roles.has(trimmed.toLowerCase())) continue;
+		if (normalized.label === "organization" && normalized.source === "legal-form" && trimmed === trimmed.toUpperCase() && LEGAL_FORM_HEADING_RE.test(trimmed)) continue;
+		if (normalized.label === "address" && trimmed.length > 40 && !POSTAL_CODE_RE.test(trimmed) && !HAS_DIGIT_RE.test(trimmed) && !ADDRESS_COMPONENTS_RE.test(trimmed) && !JURISDICTION_RE.test(trimmed)) continue;
+		if (normalized.label === "address" && normalized.source === "trigger" && !HAS_DIGIT_RE.test(trimmed) && !ADDRESS_COMPONENTS_RE.test(trimmed) && !JURISDICTION_RE.test(trimmed)) continue;
+		if (normalized.label === "address" && SIGNING_CLAUSE_ADDRESS_RE.test(trimmed)) continue;
+		filtered.push(normalized);
 	}
 	return filtered;
 };
@@ -2557,6 +2779,8 @@ const loadPersonStopwords = (ctx) => {
 const EMPTY_PERSON_STOPWORDS = /* @__PURE__ */ new Set();
 /** Sync accessor — returns empty set before init. */
 const getPersonStopwords = (ctx) => ctx.personStopwords ?? EMPTY_PERSON_STOPWORDS;
+const PERSON_CHAIN_BREAK_RE = /[!?;:]|,/u;
+const isInitialContinuationGap = (text, gap) => /^\p{Lu}$/u.test(text) && /^\.[^\S\n]{1,2}$/u.test(gap) || /^[^\S\n]{1,2}(?:\p{Lu}\.[^\S\n]{1,2})+$/u.test(gap);
 /**
 * Resolve which dictionaries to load based on country
 * and category filters, load them, and build the deny
@@ -2583,6 +2807,7 @@ const buildDenyList = async (config, ctx = defaultContext) => {
 	const ids = [...dataModule.ALL_DICTIONARY_IDS].filter((id) => {
 		const meta = dataModule.DICTIONARY_META[id];
 		if (!meta) return false;
+		if (!config.enableNameCorpus && meta.category === "Names") return false;
 		if (excludeCategories.has(meta.category)) return false;
 		if (allowedCountries === null) return true;
 		if (meta.country === null) return true;
@@ -2669,20 +2894,22 @@ const buildDenyList = async (config, ctx = defaultContext) => {
 			sourceList.push([source]);
 		}
 	};
-	for (const name of getNameCorpusFirstNames(ctx)) addNameEntry(name, "first-name");
-	for (const name of getNameCorpusSurnames(ctx)) addNameEntry(name, "surname");
-	for (const title of getNameCorpusTitles(ctx)) {
-		const norm = normalizeForSearch(title).replace(/[|\\]/g, "");
-		if (norm.length === 0) continue;
-		const lower = norm.toLowerCase();
-		const existing = patternIndex.get(lower);
-		if (existing !== void 0) {
-			if (!sourceList[existing].includes("title")) sourceList[existing].push("title");
-		} else {
-			patternIndex.set(lower, patternList.length);
-			patternList.push(norm);
-			labelList.push(["person"]);
-			sourceList.push(["title"]);
+	if (config.enableNameCorpus && !excludeCategories.has("Names")) {
+		for (const name of getNameCorpusFirstNames(ctx)) addNameEntry(name, "first-name");
+		for (const name of getNameCorpusSurnames(ctx)) addNameEntry(name, "surname");
+		for (const title of getNameCorpusTitles(ctx)) {
+			const norm = normalizeForSearch(title).replace(/[|\\]/g, "");
+			if (norm.length === 0) continue;
+			const lower = norm.toLowerCase();
+			const existing = patternIndex.get(lower);
+			if (existing !== void 0) {
+				if (!sourceList[existing].includes("title")) sourceList[existing].push("title");
+			} else {
+				patternIndex.set(lower, patternList.length);
+				patternList.push(norm);
+				labelList.push(["person"]);
+				sourceList.push(["title"]);
+			}
 		}
 	}
 	if (patternList.length === 0) return null;
@@ -2782,7 +3009,8 @@ const processDenyListMatches = (allMatches, sliceStart, sliceEnd, fullText, data
 			const prev = chain.at(-1);
 			if (!prev) break;
 			const gap = fullText.slice(prev.end, next.start);
-			if (gap.length > 4 || gap.length === 0 || gap.includes("\n") || gap.includes("	")) break;
+			const breaksOnPeriod = gap.includes(".") && !isInitialContinuationGap(prev.text, gap);
+			if (gap.length > 4 || gap.length === 0 || gap.includes("\n") || gap.includes("	") || PERSON_CHAIN_BREAK_RE.test(gap) || breaksOnPeriod) break;
 			chain.push(next);
 			j++;
 		}
@@ -2792,7 +3020,7 @@ const processDenyListMatches = (allMatches, sliceStart, sliceEnd, fullText, data
 		if (!first || !last) continue;
 		const extended = extendPersonName(fullText, first.start, last.end, ctx);
 		const score = chain.length >= 2 ? .9 : .5;
-		if (chain.length === 1 && isSentenceStart(fullText, first.start)) {
+		if (chain.length === 1) {
 			const afterEnd = last.end;
 			const rest = fullText.slice(afterEnd).trimStart();
 			if (!(rest.length > 1 && /^\p{Lu}\p{Ll}/u.test(rest))) continue;
@@ -3687,6 +3915,25 @@ const initHotwordRules = async () => {
 	return initPromise;
 };
 /**
+* Expand requested output labels with any source labels
+* that hotword rules may reclassify into them.
+*
+* Example: requesting only "date of birth" still needs
+* "date" candidates to survive until the hotword pass.
+* If rules are not initialized, or if no labels were
+* requested, returns the input labels unchanged.
+*/
+const expandLabelsForHotwordRules = (requestedLabels) => {
+	if (rules === null || requestedLabels.length === 0) return requestedLabels;
+	const requested = new Set(requestedLabels);
+	const expanded = new Set(requestedLabels);
+	for (const rule of rules) {
+		if (rule.reclassifyTo === void 0 || !requested.has(rule.reclassifyTo)) continue;
+		for (const label of rule.targetLabels) expanded.add(label);
+	}
+	return [...expanded];
+};
+/**
 * Apply hotword context rules to detected entities.
 *
 * Scans `fullText` once with a single AC automaton
@@ -3797,7 +4044,8 @@ const WORD_START_STOPS = new Set([
 	"(",
 	")",
 	"[",
-	"]"
+	"]",
+	"&"
 ]);
 /**
 * Find the word-start offset at or before `pos`.
@@ -3826,7 +4074,8 @@ const WORD_END_STOPS = new Set([
 	"(",
 	")",
 	"[",
-	"]"
+	"]",
+	"&"
 ]);
 /**
 * Find the word-end offset at or after `pos`.
@@ -4056,8 +4305,9 @@ const enforceBoundaryConsistency = (entities, fullText) => {
 //#endregion
 //#region src/build-unified-search.ts
 const buildUnifiedSearch = async (config, gazetteerEntries = [], ctx = defaultContext) => {
+	const legalFormsEnabled = isLegalFormsEnabled(config);
 	const [legalForms, triggers, denyListData, streetTypes, currencyPatterns, datePatterns, signingPatterns] = await Promise.all([
-		buildLegalFormPatterns(),
+		legalFormsEnabled ? buildLegalFormPatterns() : Promise.resolve([]),
 		config.enableTriggerPhrases ? buildTriggerPatterns() : Promise.resolve({
 			patterns: [],
 			rules: []
@@ -4288,7 +4538,7 @@ const COLON_LABELS = new Set(["ip address", "mac address"]);
 /** Strip leading/trailing whitespace and punctuation. */
 const sanitizeEntities = (entities) => entities.flatMap((e) => {
 	const strip = COLON_LABELS.has(e.label) ? /[\s,;]+/ : /[\s:,;]+/;
-	const leadTrimmed = e.text.replace(new RegExp(`^${strip.source}`, strip.flags), "");
+	const leadTrimmed = e.text.replace(/^(?:\.\s)+/, "").replace(new RegExp(`^${strip.source}`, strip.flags), "");
 	const lead = e.text.length - leadTrimmed.length;
 	const cleaned = leadTrimmed.replace(new RegExp(`${strip.source}$`, strip.flags), "");
 	if (cleaned.length === 0) return [];
@@ -4345,12 +4595,24 @@ const extendMonetaryAmountWords = (entities, fullText, re) => entities.map((e) =
 		text: fullText.slice(e.start, newEnd)
 	};
 });
+const createAllowedLabelSetFromLabels = (labels) => labels.length > 0 ? new Set(labels) : null;
+const createAllowedLabelSet = (config) => createAllowedLabelSetFromLabels(config.labels);
+const filterAllowedLabels = (entities, allowedLabels) => {
+	if (!allowedLabels) return entities;
+	return entities.filter((e) => allowedLabels.has(e.label));
+};
+const labelIsAllowed = (label, allowedLabels) => !allowedLabels || allowedLabels.has(label);
+const getRequestedNerLabels = (config, expandForHotwords = false) => {
+	const labels = config.labels.length > 0 ? config.labels : DEFAULT_ENTITY_LABELS;
+	return expandForHotwords ? expandLabelsForHotwordRules(labels) : labels;
+};
 const checkAbort = (signal) => {
 	if (signal?.aborted) throw new DOMException("Pipeline aborted", "AbortError");
 };
 const configKey = (config, gazetteerEntries) => {
+	const legalFormsEnabled = isLegalFormsEnabled(config);
 	const gazFingerprint = config.enableGazetteer && gazetteerEntries.length > 0 ? gazetteerEntries.map((e) => `${e.id}:${e.canonical}:${e.label}:${[...e.variants].sort().join(",")}`).toSorted().join(";") : "";
-	return `${config.enableDenyList}:${config.enableTriggerPhrases}:${config.denyListCountries?.toSorted().join(",") ?? ""}:${config.denyListRegions?.toSorted().join(",") ?? ""}:${config.denyListExcludeCategories?.toSorted().join(",") ?? ""}:${config.enableGazetteer}:${gazFingerprint}`;
+	return `${config.enableDenyList}:${config.enableTriggerPhrases}:${legalFormsEnabled}:${config.enableNameCorpus}:${config.denyListCountries?.toSorted().join(",") ?? ""}:${config.denyListRegions?.toSorted().join(",") ?? ""}:${config.denyListExcludeCategories?.toSorted().join(",") ?? ""}:${config.enableGazetteer}:${gazFingerprint}`;
 };
 /**
 * Get or build a cached search instance. Cache state
@@ -4387,6 +4649,8 @@ const getCachedSearch = async (config, gazetteerEntries, ctx) => {
 const runPipeline = async (options) => {
 	const { fullText, config, gazetteerEntries, nerInference = null, onProgress, cachedSearch, signal, context } = options;
 	const ctx = context ?? defaultContext;
+	const allowedLabels = createAllowedLabelSet(config);
+	const legalFormsEnabled = isLegalFormsEnabled(config);
 	const log = (step, detail) => {
 		onProgress?.(step, detail);
 	};
@@ -4427,44 +4691,57 @@ const runPipeline = async (options) => {
 		if (zones.length > 0) log("zones", [...new Set(zones.map((z) => z.zone))].join(", "));
 	}
 	checkAbort(signal);
+	const hotwordsActive = enableHotwords && hotwordInitOk;
+	const preHotwordAllowedLabels = hotwordsActive ? createAllowedLabelSetFromLabels(expandLabelsForHotwordRules(config.labels)) : allowedLabels;
 	const search = cachedSearch ?? await getCachedSearch(config, gazetteerEntries, ctx);
 	checkAbort(signal);
 	const { regexMatches, literalMatches } = runUnifiedSearch(search, fullText);
 	const { slices } = search;
-	const regexEntities = config.enableRegex ? processRegexMatches(regexMatches, slices.regex.start, slices.regex.end, search.regexMeta) : [];
+	const rawRegexEntities = config.enableRegex ? processRegexMatches(regexMatches, slices.regex.start, slices.regex.end, search.regexMeta) : [];
+	const regexEntities = filterAllowedLabels(rawRegexEntities, preHotwordAllowedLabels);
 	if (regexEntities.length > 0) log("regex", `${regexEntities.length} matches`);
-	const legalFormEntities = processLegalFormMatches(regexMatches, slices.legalForms.start, slices.legalForms.end, fullText);
+	const rawLegalFormEntities = legalFormsEnabled ? processLegalFormMatches(regexMatches, slices.legalForms.start, slices.legalForms.end, fullText) : [];
+	const legalFormEntities = filterAllowedLabels(rawLegalFormEntities, preHotwordAllowedLabels);
 	if (legalFormEntities.length > 0) log("legal-forms", `${legalFormEntities.length} matches`);
-	const triggerEntities = config.enableTriggerPhrases ? processTriggerMatches(regexMatches, slices.triggers.start, slices.triggers.end, fullText, search.triggerRules) : [];
+	const rawTriggerEntities = config.enableTriggerPhrases ? processTriggerMatches(regexMatches, slices.triggers.start, slices.triggers.end, fullText, search.triggerRules) : [];
+	const triggerEntities = filterAllowedLabels(rawTriggerEntities, preHotwordAllowedLabels);
 	if (triggerEntities.length > 0) log("trigger-phrases", `${triggerEntities.length} matches`);
 	checkAbort(signal);
+	let rawNameCorpusEntities = [];
 	let nameCorpusEntities = [];
 	if (config.enableNameCorpus && !config.enableDenyList) {
 		await initNameCorpus(ctx);
 		checkAbort(signal);
-		nameCorpusEntities = detectNameCorpus(fullText, ctx);
+		rawNameCorpusEntities = detectNameCorpus(fullText, ctx);
+		nameCorpusEntities = filterAllowedLabels(rawNameCorpusEntities, preHotwordAllowedLabels);
 		log("name-corpus", `${nameCorpusEntities.length} matches`);
 	}
-	const denyListEntities = config.enableDenyList && search.denyListData ? processDenyListMatches(literalMatches, slices.denyList.start, slices.denyList.end, fullText, search.denyListData, ctx) : [];
+	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", `${denyListEntities.length} matches`);
-	const gazetteerEntities = config.enableGazetteer && search.gazetteerData ? processGazetteerMatches(literalMatches, slices.gazetteer.start, slices.gazetteer.end, fullText, search.gazetteerData) : [];
+	const rawGazetteerEntities = config.enableGazetteer && search.gazetteerData ? processGazetteerMatches(literalMatches, slices.gazetteer.start, slices.gazetteer.end, fullText, search.gazetteerData) : [];
+	const gazetteerEntities = filterAllowedLabels(rawGazetteerEntities, preHotwordAllowedLabels);
 	if (gazetteerEntities.length > 0) log("gazetteer", `${gazetteerEntities.length} matches`);
 	checkAbort(signal);
+	const ruleContextEntities = [
+		...rawTriggerEntities,
+		...rawRegexEntities,
+		...rawLegalFormEntities,
+		...rawNameCorpusEntities,
+		...rawDenyListEntities,
+		...rawGazetteerEntities
+	];
+	let rawNerEntities = [];
 	let nerEntities = [];
 	if (config.enableNer && nerInference) {
-		const maskResult = maskDetectedSpans(fullText, [
-			...triggerEntities,
-			...regexEntities,
-			...legalFormEntities,
-			...nameCorpusEntities,
-			...denyListEntities,
-			...gazetteerEntities
-		]);
+		const maskResult = maskDetectedSpans(fullText, ruleContextEntities);
 		log("ner", "running inference...");
-		const rawNer = await nerInference(maskResult.maskedText, config.labels, config.threshold, signal);
-		nerEntities = unmaskNerEntities(rawNer, maskResult, fullText);
-		const dropped = rawNer.length - nerEntities.length;
-		log("ner", `${nerEntities.length} entities` + (dropped > 0 ? ` (${dropped} masked)` : ""));
+		const rawNer = await nerInference(maskResult.maskedText, [...getRequestedNerLabels(config, hotwordsActive)], config.threshold, signal);
+		rawNerEntities = unmaskNerEntities(rawNer, maskResult, fullText);
+		nerEntities = filterAllowedLabels(rawNerEntities, preHotwordAllowedLabels);
+		const masked = rawNer.length - rawNerEntities.length;
+		const labelFiltered = rawNerEntities.length - nerEntities.length;
+		log("ner", `${nerEntities.length} entities` + (masked > 0 ? ` (${masked} masked)` : "") + (labelFiltered > 0 ? ` (${labelFiltered} label-filtered)` : ""));
 	}
 	checkAbort(signal);
 	const preAddressEntities = [
@@ -4476,23 +4753,23 @@ const runPipeline = async (options) => {
 		...gazetteerEntities,
 		...nerEntities
 	];
-	const addressSeedEntities = await processAddressSeeds(literalMatches, slices.streetTypes.start, slices.streetTypes.end, fullText, preAddressEntities);
+	const addressSeedEntities = labelIsAllowed("address", allowedLabels) ? await processAddressSeeds(literalMatches, slices.streetTypes.start, slices.streetTypes.end, fullText, [...ruleContextEntities, ...rawNerEntities]) : [];
 	if (addressSeedEntities.length > 0) log("address-seeds", `${addressSeedEntities.length} expanded`);
 	checkAbort(signal);
 	const zoneAdjusted = applyZoneAdjustments([...preAddressEntities, ...addressSeedEntities], zones);
-	const preBoostEntities = enableHotwords && hotwordInitOk ? applyHotwordRules(zoneAdjusted, fullText) : zoneAdjusted;
+	const preBoostEntities = hotwordsActive ? filterAllowedLabels(applyHotwordRules(zoneAdjusted, fullText), allowedLabels) : zoneAdjusted;
 	let allEntities;
 	if (config.enableConfidenceBoost) {
 		allEntities = boostNearMissEntities(preBoostEntities, config.threshold);
 		const boosted = allEntities.length - preBoostEntities.filter((e) => e.score >= config.threshold).length;
 		if (boosted > 0) log("confidence-boost", `${boosted} near-miss promoted`);
 	} else allEntities = preBoostEntities.filter((e) => e.score >= config.threshold);
-	const streetPatterns = detectStreetPatternsNearAddresses(fullText, allEntities);
+	const streetPatterns = labelIsAllowed("address", allowedLabels) ? detectStreetPatternsNearAddresses(fullText, allEntities) : [];
 	if (streetPatterns.length > 0) {
 		allEntities = [...allEntities, ...streetPatterns];
 		log("street-context", `${streetPatterns.length} street patterns near addresses`);
 	}
-	const orphanStreets = detectOrphanStreetLines(fullText, allEntities);
+	const orphanStreets = labelIsAllowed("address", allowedLabels) ? detectOrphanStreetLines(fullText, allEntities) : [];
 	if (orphanStreets.length > 0) {
 		allEntities = [...allEntities, ...orphanStreets];
 		log("orphan-streets", `${orphanStreets.length} header street lines`);
@@ -4503,14 +4780,14 @@ const runPipeline = async (options) => {
 	const consistent = enforceBoundaryConsistency(mergedExtended, fullText);
 	if (consistent.length < mergedExtended.length) log("boundary", `${mergedExtended.length - consistent.length} consolidated`);
 	let postOrgEntities = consistent;
-	if (config.enableCoreference) {
+	if (config.enableCoreference && labelIsAllowed("organization", allowedLabels)) {
 		const thresholded = propagateOrgNames(consistent, fullText).filter((e) => e.score >= config.threshold);
 		if (thresholded.length > 0) {
 			postOrgEntities = mergeAndDedup(consistent, thresholded);
 			log("org-propagation", `${thresholded.length} base names`);
 		}
 	}
-	const merged = filterFalsePositives(postOrgEntities, ctx);
+	const merged = filterFalsePositives(postOrgEntities, ctx, fullText);
 	if (merged.length < postOrgEntities.length) log("filter", `removed ${postOrgEntities.length - merged.length} FPs`);
 	checkAbort(signal);
 	ctx.corefSourceMap.clear();
@@ -4521,11 +4798,11 @@ const runPipeline = async (options) => {
 			const corefSpans = findCoreferenceSpans(fullText, terms, ctx);
 			if (corefSpans.length > 0) {
 				log("coreference-rescan", `${corefSpans.length} aliases`);
-				return sanitizeEntities(filterFalsePositives(enforceBoundaryConsistency(mergeAndDedup(merged, corefSpans), fullText), ctx));
+				return sanitizeEntities(filterAllowedLabels(filterFalsePositives(enforceBoundaryConsistency(mergeAndDedup(merged, corefSpans), fullText), ctx, fullText), allowedLabels));
 			}
 		}
 	}
-	return sanitizeEntities(merged);
+	return sanitizeEntities(filterAllowedLabels(merged, allowedLabels));
 };
 const OPERATOR_REGISTRY = {
 	replace: {