@stll/anonymize-wasm 1.1.1 → 1.3.0

package/dist/wasm.d.mts

@@ -35,6 +35,7 @@ type Entity = {
   text: string;
   score: number;
   source: DetectionSource;
+  sourceDetail?: "custom-deny-list" | "custom-regex";
 };
 /**
  * Entity after human review. Extends the base Entity
@@ -61,6 +62,16 @@ type GazetteerEntry = {
 /** Extraction strategy — closed discriminated union. */
 type TriggerStrategy = {
   type: "to-next-comma";
+  /**
+   * Optional list of lowercase keywords that terminate
+   * the value scan, in addition to commas/newlines. Useful
+   * for triggers like court names that may continue past
+   * a missing comma into adjacent clause text ("Městským
+   * soudem v Praze dne 1. 1. 2020"); listing `"dne"` here
+   * stops the scan at the date boundary. Matched on a
+   * word-boundary, case-insensitive.
+   */
+  stopWords?: string[];
 } | {
   type: "to-end-of-line";
 } | {
@@ -71,6 +82,19 @@ type TriggerStrategy = {
 } | {
   type: "address";
   maxChars?: number;
+} | {
+  /**
+   * Extract the first regex match in the value text.
+   * Useful for shape-bounded values that follow a
+   * label on the same line as other fields, where
+   * `to-end-of-line` would over-capture. The pattern
+   * is anchored to the start of the (already
+   * leading-whitespace-stripped) value, so use
+   * `(?:.*?)` prefix only when intentional.
+   */
+  type: "match-pattern";
+  pattern: string;
+  flags?: string;
 };
 /** Validation rules — closed discriminated union. */
 type TriggerValidation = {
@@ -89,7 +113,21 @@ type TriggerValidation = {
   type: "matches-pattern";
   pattern: string;
   flags?: string;
+}
+/**
+ * Run a named stdnum validator (checksum + length)
+ * against the captured value. Keeps the trigger
+ * path symmetrical with the formatted-regex
+ * detectors so e.g. `CPF nº 00000000000` does not
+ * survive as a tax-ID entity.
+ */
+| {
+  type: "valid-id";
+  validator: ValidIdValidator;
 };
+/** Built-in stdnum validators that can be referenced
+ *  by `valid-id` validations. */
+type ValidIdValidator = "br.cpf" | "br.cnpj";
 /** Auto-generated trigger variants — closed set. */
 type TriggerExtension = "add-colon" | "add-trailing-space" | "add-colon-space" | "normalize-spaces";
 /** V2 trigger config entry (JSON shape). */
@@ -123,6 +161,9 @@ type CompiledValidation = {
 } | {
   type: "matches-pattern";
   re: RegExp;
+} | {
+  type: "valid-id";
+  check: (value: string) => boolean;
 };
 /**
  * Runtime rule — one per trigger string after
@@ -185,6 +226,27 @@ type DictionaryMeta = {
   category: DenyListCategory;
   country: string | null;
 };
+/**
+ * Caller-supplied exact terms for deny-list matching.
+ * These entries are merged with the published deny-list
+ * dictionaries when `enableDenyList` is enabled.
+ */
+type CustomDenyListEntry = {
+  value: string;
+  label: string;
+  variants?: readonly string[];
+};
+/**
+ * Caller-supplied regex detector. The pattern is passed
+ * to the underlying text-search regex engine, so use its
+ * supported regex syntax. Inline flags such as `(?i)` are
+ * accepted when supported by that engine.
+ */
+type CustomRegexPattern = {
+  pattern: string;
+  label: string;
+  score?: number;
+};
 /**
  * Pre-loaded dictionary data for dependency injection.
  * Consumers that want name/city/deny-list detection
@@ -248,6 +310,16 @@ type PipelineConfig = {
   denyListCountries?: string[];
   denyListRegions?: string[];
   denyListExcludeCategories?: string[];
+  /**
+   * Caller-owned exact terms to match through the
+   * deny-list layer. Requires `enableDenyList: true`.
+   */
+  customDenyList?: readonly CustomDenyListEntry[];
+  /**
+   * Caller-owned regex detectors. Requires
+   * `enableRegex: true`.
+   */
+  customRegexes?: readonly CustomRegexPattern[];
   enableGazetteer: boolean;
   enableNer: boolean;
   enableConfidenceBoost: boolean;
@@ -278,12 +350,13 @@ type PipelineConfig = {
  * every pipeline run and never persisted to the database.
  * Renaming a label here requires no migration.
  */
-declare const DEFAULT_ENTITY_LABELS: readonly ["person", "organization", "phone number", "address", "email address", "date", "date of birth", "bank account number", "iban", "tax identification number", "identity card number", "registration number", "credit card number", "passport number", "monetary amount", "land parcel"];
+declare const DEFAULT_ENTITY_LABELS: readonly ["person", "organization", "phone number", "address", "email address", "date", "date of birth", "bank account number", "iban", "tax identification number", "identity card number", "birth number", "national identification number", "social security number", "registration number", "credit card number", "passport number", "monetary amount", "land parcel", "misc"];
 //#endregion
 //#region src/detectors/regex.d.ts
 type RegexMeta = {
   label: string;
-  score: number; /** Post-match stdnum validator for confirmation. */
+  score: number;
+  sourceDetail?: Entity["sourceDetail"]; /** Post-match stdnum validator for confirmation. */
   validator?: Validator;
 };
 /** Flat pattern array for text-search. */
@@ -320,15 +393,16 @@ declare const CURRENCY_PATTERN_META: Readonly<RegexMeta>;
 declare const processRegexMatches: (allMatches: Match[], sliceStart: number, sliceEnd: number, meta_: readonly RegexMeta[]) => Entity[];
 //#endregion
 //#region src/detectors/deny-list.d.ts
-type DenyListConfig = Pick<PipelineConfig, "enableDenyList" | "enableNameCorpus" | "denyListCountries" | "denyListRegions" | "denyListExcludeCategories" | "dictionaries">;
+type DenyListConfig = Pick<PipelineConfig, "enableDenyList" | "enableNameCorpus" | "denyListCountries" | "denyListRegions" | "denyListExcludeCategories" | "customDenyList" | "dictionaries">;
 /**
  * Source tag for each pattern in the automaton.
  * "deny-list" = standard deny list entry
+ * "custom-deny-list" = caller-owned exact term
  * "first-name" = name corpus first name
  * "surname" = name corpus surname
  * "title" = academic/professional title
  */
-type PatternSource = "deny-list" | "first-name" | "surname" | "title";
+type PatternSource = "deny-list" | "custom-deny-list" | "first-name" | "surname" | "title";
 /**
  * Pre-built deny list data. Constructed once by
  * `buildDenyList`, reused across `processDenyListMatches`
@@ -342,7 +416,8 @@ type DenyListData = {
    * appears in multiple dictionaries (e.g., "Denver"
    * is both a person name and a city name).
    */
-  labels: string[][]; /** Maps pattern index → original pattern text. */
+  labels: string[][]; /** Maps pattern index → labels contributed by custom entries. */
+  customLabels: string[][]; /** Maps pattern index → original pattern text. */
   originals: string[]; /** Maps pattern index → source types (plural). */
   sources: PatternSource[][];
 };
@@ -396,10 +471,12 @@ type GazetteerData = {
   isFuzzy: boolean[];
 };
 type UnifiedSearchInstance = {
-  /** Regex + triggers + legal-forms. */tsRegex: TextSearch; /** Deny-list + street-types + gazetteer. */
+  /** Regex + triggers + legal-forms. */tsRegex: TextSearch; /** Caller-owned custom regexes, isolated for overlap preservation. */
+  tsCustomRegex: TextSearch; /** Deny-list + street-types + gazetteer. */
   tsLiterals: TextSearch;
   slices: {
     regex: PatternSlice;
+    customRegex: PatternSlice;
     legalForms: PatternSlice;
     triggers: PatternSlice;
     denyList: PatternSlice;
@@ -407,6 +484,7 @@ type UnifiedSearchInstance = {
     gazetteer: PatternSlice;
   };
   regexMeta: readonly RegexMeta[];
+  customRegexMeta: readonly RegexMeta[];
   triggerRules: readonly TriggerRule[];
   denyListData: DenyListData | null;
   gazetteerData: GazetteerData | null;
@@ -464,7 +542,9 @@ type PipelineContext = {
   allowList: ReadonlySet<string> | null;
   allowListPromise: Promise<ReadonlySet<string>> | null;
   personStopwords: ReadonlySet<string> | null;
-  personStopwordsPromise: Promise<ReadonlySet<string>> | null; /** First-name exclusions for stopword filtering. */
+  personStopwordsPromise: Promise<ReadonlySet<string>> | null;
+  addressStopwords: ReadonlySet<string> | null;
+  addressStopwordsPromise: Promise<ReadonlySet<string>> | null; /** First-name exclusions for stopword filtering. */
   firstNameExclusions: ReadonlySet<string> | null;
   firstNameExclusionCorpusLen: number;
   genericRoles: ReadonlySet<string> | null;
@@ -589,6 +669,7 @@ declare const DEFAULT_OPERATOR_CONFIG: OperatorConfig;
 declare const resolveOperator: (config: OperatorConfig, label: string) => OperatorType;
 //#endregion
 //#region src/detectors/legal-forms.d.ts
+declare const warmLegalRoleHeads: () => Promise<void>;
 /**
  * Build legal form regex pattern strings.
  * Returns an array of regex strings for the unified
@@ -600,6 +681,14 @@ declare const buildLegalFormPatterns: () => Promise<string[]>;
  * Process legal form matches from the unified search.
  * Receives all matches; filters to the legal forms
  * slice via sliceStart/sliceEnd.
+ *
+ * The role-head trimming step reads per-language data from
+ * a cache that `runPipeline` warms via `warmLegalRoleHeads()`
+ * before calling this. Callers that invoke
+ * `processLegalFormMatches` directly (without going through
+ * `runPipeline`) must `await warmLegalRoleHeads()` first;
+ * otherwise the trim falls back to a no-op and sentence-
+ * fragment fixes do not apply.
  */
 declare const processLegalFormMatches: (allMatches: Match[], sliceStart: number, sliceEnd: number, fullText?: string) => Entity[];
 //#endregion
@@ -739,6 +828,7 @@ declare const detectNameCorpus: (fullText: string, ctx?: PipelineContext) => Ent
 //#region src/unified-search.d.ts
 type UnifiedResult = {
   /** All matches from both instances combined. */regexMatches: Match[];
+  customRegexMatches: Match[];
   literalMatches: Match[];
 };
 declare const runUnifiedSearch: (instance: UnifiedSearchInstance, fullText: string) => UnifiedResult;
@@ -780,6 +870,8 @@ type CountryCode = NonNullRegion[keyof NonNullRegion][number];
 declare const resolveCountries: (regions?: string[], countries?: string[]) => Set<string> | null;
 //#endregion
 //#region src/filters/false-positives.d.ts
+/** Ensure street-type vocabulary is loaded. */
+declare const initAddressComponents: () => Promise<void>;
 /**
  * Filter out entities that are likely false positives:
  * template placeholders, clause/section numbers,
@@ -996,5 +1088,5 @@ declare const levenshtein: (rawA: string, rawB: string) => number;
  */
 declare const normalizeForSearch: (text: string) => string;
 //#endregion
-export { type AnonymisationOperator, CURRENCY_PATTERN_META, type CountryCode, DATE_PATTERN_META, DEFAULT_ENTITY_LABELS, DEFAULT_OPERATOR_CONFIG, DETECTION_SOURCES, DETECTOR_PRIORITY, type DefinitionPattern, type DenyListCategory, type DenyListData, type DetectionSource, type Dictionaries, type DictionaryMeta, type DocumentZone, type Entity, type EntityResult, type GazetteerData, type GazetteerEntry, type HotwordRule, type NameCorpusData, type NerInferenceFn, OPERATOR_REGISTRY, OPERATOR_TYPES, type OperatorConfig, type OperatorType, type PipelineConfig, type PipelineContext, type PipelineOptions, REGEX_META, REGEX_PATTERNS, REGIONS, type RawInferenceResult, type RedactionResult, type RegexMeta, type RegionId, type ReviewDecision, type ReviewedEntity, type TriggerExtension, type TriggerGroupConfig, type TriggerRule, type TriggerStrategy, type TriggerValidation, type UnifiedResult, type UnifiedSearchInstance, ZONE_SCORE_ADJUSTMENTS, type ZoneSpan, 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, initHotwordRules, initNameCorpus, initZoneClassifier, levenshtein, mergeAndDedup, mergeChunkEntities, normalizeForSearch, prepareBatch, processAddressSeeds, processDenyListMatches, processGazetteerMatches, processLegalFormMatches, processRegexMatches, processTriggerMatches, propagateOrgNames, redactText, resolveCountries, resolveOperator, runPipeline, runUnifiedSearch, sanitizeEntities, tokenizeText };
+export { type AnonymisationOperator, CURRENCY_PATTERN_META, type CountryCode, type CustomDenyListEntry, type CustomRegexPattern, DATE_PATTERN_META, DEFAULT_ENTITY_LABELS, DEFAULT_OPERATOR_CONFIG, DETECTION_SOURCES, DETECTOR_PRIORITY, type DefinitionPattern, type DenyListCategory, type DenyListData, type DetectionSource, type Dictionaries, type DictionaryMeta, type DocumentZone, type Entity, type EntityResult, type GazetteerData, type GazetteerEntry, type HotwordRule, type NameCorpusData, type NerInferenceFn, OPERATOR_REGISTRY, OPERATOR_TYPES, type OperatorConfig, type OperatorType, type PipelineConfig, type PipelineContext, type PipelineOptions, REGEX_META, REGEX_PATTERNS, REGIONS, type RawInferenceResult, type RedactionResult, type RegexMeta, type RegionId, type ReviewDecision, type ReviewedEntity, type TriggerExtension, type TriggerGroupConfig, type TriggerRule, type TriggerStrategy, type TriggerValidation, type UnifiedResult, type UnifiedSearchInstance, ZONE_SCORE_ADJUSTMENTS, type ZoneSpan, 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, processAddressSeeds, processDenyListMatches, processGazetteerMatches, processLegalFormMatches, processRegexMatches, processTriggerMatches, propagateOrgNames, redactText, resolveCountries, resolveOperator, runPipeline, runUnifiedSearch, sanitizeEntities, tokenizeText, warmLegalRoleHeads };
 //# sourceMappingURL=wasm.d.mts.map