euredact 0.6.0 → 0.7.0

package/dist/esm/rules/validators.d.ts

@@ -65,7 +65,21 @@ export declare function validateSpanishDni(candidate: string): boolean;
 export declare function validateSpanishNie(candidate: string): boolean;
 export declare function validatePortugueseNif(candidate: string): boolean;
 export declare function validatePolishPesel(candidate: string): boolean;
+/** Polish domestic account number (NRB): the PL IBAN without its "PL", so the IBAN's mod-97 decides it (rules-engine#93). */
+export declare function validatePolishNrb(candidate: string): boolean;
 export declare function validatePolishNip(candidate: string): boolean;
+/**
+ * Polish identity card (dowód osobisty): 3 letters + 6 digits, the first digit
+ * a check digit. Letters A=10 … Z=35, weights 7,3,1,9,7,3,1,7,3: the weighted
+ * sum of all nine is 0 mod 10 exactly when the check digit is right (rules-engine#75).
+ */
+export declare function validatePolishIdCard(candidate: string): boolean;
+/**
+ * Polish REGON: 9 digits, or 14 for a local unit, each with a mod-11 check (a
+ * remainder of 10 counts as 0). A 14-digit REGON begins with its parent's valid
+ * 9-digit REGON (rules-engine#76).
+ */
+export declare function validatePolishRegon(candidate: string): boolean;
 export declare function validateCzechBirthNumber(candidate: string): boolean;
 export declare function validateRomanianCnp(candidate: string): boolean;
 export declare function validateHungarianTaj(candidate: string): boolean;

package/dist/esm/rules/validators.js

@@ -492,6 +492,13 @@ export function validatePolishPesel(candidate) {
         total += d[i] * weights[i];
     return d[10] === (10 - (total % 10)) % 10;
 }
+/** Polish domestic account number (NRB): the PL IBAN without its "PL", so the IBAN's mod-97 decides it (rules-engine#93). */
+export function validatePolishNrb(candidate) {
+    const c = candidate.replace(/[\s\-]/g, "");
+    if (!/^\d{26}$/.test(c))
+        return false;
+    return validateIban("PL" + c);
+}
 export function validatePolishNip(candidate) {
     const c = candidate.replace(/[\s\-]/g, "");
     if (c.length !== 10 || !/^\d{10}$/.test(c))
@@ -503,6 +510,43 @@ export function validatePolishNip(candidate) {
         total += d[i] * weights[i];
     return total % 11 === d[9];
 }
+/**
+ * Polish identity card (dowód osobisty): 3 letters + 6 digits, the first digit
+ * a check digit. Letters A=10 … Z=35, weights 7,3,1,9,7,3,1,7,3: the weighted
+ * sum of all nine is 0 mod 10 exactly when the check digit is right (rules-engine#75).
+ */
+export function validatePolishIdCard(candidate) {
+    const c = candidate.replace(/\s/g, "").toUpperCase();
+    if (!/^[A-Z]{3}\d{6}$/.test(c))
+        return false;
+    const weights = [7, 3, 1, 9, 7, 3, 1, 7, 3];
+    let total = 0;
+    for (let i = 0; i < 9; i++) {
+        const ch = c[i];
+        total += (/[A-Z]/.test(ch) ? ch.charCodeAt(0) - 55 : Number(ch)) * weights[i];
+    }
+    return total % 10 === 0;
+}
+/**
+ * Polish REGON: 9 digits, or 14 for a local unit, each with a mod-11 check (a
+ * remainder of 10 counts as 0). A 14-digit REGON begins with its parent's valid
+ * 9-digit REGON (rules-engine#76).
+ */
+export function validatePolishRegon(candidate) {
+    const c = candidate.replace(/[\s\-]/g, "");
+    if (!/^\d+$/.test(c) || (c.length !== 9 && c.length !== 14))
+        return false;
+    const check = (digits, weights) => {
+        let total = 0;
+        for (let i = 0; i < weights.length; i++)
+            total += Number(digits[i]) * weights[i];
+        const rest = total % 11;
+        return (rest === 10 ? 0 : rest) === Number(digits[weights.length]);
+    };
+    if (!check(c.slice(0, 9), [8, 9, 2, 3, 4, 5, 6, 7]))
+        return false;
+    return c.length === 9 || check(c, [2, 4, 8, 5, 0, 9, 7, 3, 6, 1, 2, 4, 8]);
+}
 // Days per month for the rodné číslo date check. February is 29 because the
 // century is not recoverable from a two-digit year, so a leap year cannot be
 // ruled out -- being permissive by one day is the right direction here.
@@ -753,6 +797,9 @@ export const VALIDATORS = {
     portuguese_nif: validatePortugueseNif,
     polish_pesel: validatePolishPesel,
     polish_nip: validatePolishNip,
+    polish_nrb: validatePolishNrb,
+    polish_id_card: validatePolishIdCard,
+    polish_regon: validatePolishRegon,
     czech_birth_number: validateCzechBirthNumber,
     romanian_cnp: validateRomanianCnp,
     hungarian_taj: validateHungarianTaj,

package/dist/esm/sdk.d.ts

@@ -22,11 +22,51 @@ type LabelFor = (det: Detection, slice: string) => string;
  */
 export declare function applyReplacements(text: string, detections: Detection[], labelFor: LabelFor): string;
 /**
- * Characters a token suffix is drawn from. No vowels, so a suffix never spells
- * a word; no 0/1/I/O, so it survives being read back by a person; no
- * underscore, so the type prefix stays unambiguous.
+ * One label in the text sent to the cloud tier: where it sits in the masked
+ * text, and the extent of the original it replaced —
+ * `[maskedStart, maskedEnd, start, end]`.
  */
-export declare const TOKEN_ALPHABET = "ABCDEFGHJKLMNPQRSTUVWXYZ23456789";
+export type WireLabel = [number, number, number, number];
+/**
+ * Bracket-mask `text` and record where each label landed.
+ *
+ * Returns what is sent to the cloud tier, plus one `WireLabel` per label.
+ * Everything between labels is copied unchanged, so the labels are all it
+ * takes to carry an offset from the masked text back to the original.
+ *
+ * Always `[TYPE]`, whatever label scheme the caller asked for. That is the
+ * syntax the model was trained on, where a placeholder means "already handled,
+ * do not list it". A token or a numbered label would be read as ordinary text,
+ * so the format on the wire is a model contract and not an output option.
+ */
+/** @internal Shared with cloud/batches.ts. */
+export declare function maskForCloud(text: string, detections: Detection[]): [string, WireLabel[]];
+/**
+ * Carry the service's spans from the masked text onto the original.
+ *
+ * The service only ever saw `masked`, so its offsets index that. An offset in
+ * copied text shifts by whatever the labels before it added or removed. An
+ * offset inside a label has no counterpart in the original, so it snaps
+ * outward to the edge of the value the label replaced: an address the model
+ * reports around `[POSTAL_CODE]` covers the postal code, never half of it.
+ * Over-masking is the safe direction.
+ *
+ * A span lying wholly inside one label describes the label, not the document.
+ * It is dropped; the local detection behind the label stands.
+ *
+ * Throws `Unplaceable` when a span does not match the text that was sent. Such
+ * a span cannot be placed, and masking where it points would cover the wrong
+ * characters and leave the right ones in the clear.
+ */
+/** @internal Shared with cloud/batches.ts. */
+export declare function ontoOriginal(spans: Detection[], masked: string, text: string, labels: WireLabel[], Unplaceable: new (message: string) => Error): Detection[];
+/**
+ * Characters a token suffix is drawn from. No vowels (and no Y), so a suffix
+ * never spells a word; no 0/1/I/O, so it survives being read back by a person;
+ * no underscore, so the type prefix stays unambiguous. A, E, U and Y were in it
+ * until rules-engine#55, against this comment.
+ */
+export declare const TOKEN_ALPHABET = "BCDFGHJKLMNPQRSTVWXZ23456789";
 export declare const TOKEN_SUFFIX_LENGTH = 4;
 /**
  * Maps each distinct PII value in one call to a reversible token.
@@ -106,6 +146,13 @@ export interface RedactOptions {
     allowlistDomains?: string[] | null;
     detectDates?: boolean;
     cache?: boolean;
+    /**
+     * `mode: "cloud"` only. Sent as the request's `Idempotency-Key`: the same key
+     * returns the same job instead of running (and billing) the document again,
+     * which is what a caller retrying after its own timeout needs. One is
+     * generated per call when omitted.
+     */
+    idempotencyKey?: string;
 }
 export declare class EuRedact {
     private engine;

package/dist/esm/sdk.js

@@ -1,4 +1,4 @@
-import { randomInt } from "node:crypto";
+import { randomIndex } from "./platform.js";
 import { ResultCache } from "./cache.js";
 import { normalize, mapOffsetToOriginal } from "./normalizer.js";
 import { RuleEngine } from "./rules/engine.js";
@@ -60,7 +60,8 @@ export function applyReplacements(text, detections, labelFor) {
  * do not list it". A token or a numbered label would be read as ordinary text,
  * so the format on the wire is a model contract and not an output option.
  */
-function maskForCloud(text, detections) {
+/** @internal Shared with cloud/batches.ts. */
+export function maskForCloud(text, detections) {
     const parts = [];
     const labels = [];
     let pos = 0;
@@ -110,7 +111,8 @@ function labelBefore(labels, offset, strict) {
  * a span cannot be placed, and masking where it points would cover the wrong
  * characters and leave the right ones in the clear.
  */
-function ontoOriginal(spans, masked, text, labels, Unplaceable) {
+/** @internal Shared with cloud/batches.ts. */
+export function ontoOriginal(spans, masked, text, labels, Unplaceable) {
     const placed = [];
     for (const span of spans) {
         if (!(0 <= span.start && span.start <= span.end && span.end <= masked.length) ||
@@ -139,11 +141,12 @@ function ontoOriginal(spans, masked, text, labels, Unplaceable) {
     return placed;
 }
 /**
- * Characters a token suffix is drawn from. No vowels, so a suffix never spells
- * a word; no 0/1/I/O, so it survives being read back by a person; no
- * underscore, so the type prefix stays unambiguous.
+ * Characters a token suffix is drawn from. No vowels (and no Y), so a suffix
+ * never spells a word; no 0/1/I/O, so it survives being read back by a person;
+ * no underscore, so the type prefix stays unambiguous. A, E, U and Y were in it
+ * until rules-engine#55, against this comment.
  */
-export const TOKEN_ALPHABET = "ABCDEFGHJKLMNPQRSTUVWXYZ23456789";
+export const TOKEN_ALPHABET = "BCDFGHJKLMNPQRSTVWXZ23456789";
 export const TOKEN_SUFFIX_LENGTH = 4;
 const TOKEN_MAX_DRAWS = 100;
 function escapeRegExp(s) {
@@ -191,7 +194,7 @@ export class TokenMapper {
                 }
                 let suffix = "";
                 for (let i = 0; i < TOKEN_SUFFIX_LENGTH; i++) {
-                    suffix += TOKEN_ALPHABET[randomInt(TOKEN_ALPHABET.length)];
+                    suffix += TOKEN_ALPHABET[randomIndex(TOKEN_ALPHABET.length)];
                 }
                 token = prefix + suffix;
             } while (this.taken.has(token));
@@ -527,7 +530,9 @@ export class EuRedact {
         const [masked, labels] = maskForCloud(text, local.detections);
         // The client converts the service's code-point offsets to UTF-16 units of
         // the text it sent, so everything from here on is in one unit.
-        const remote = await client.redact(masked, { country: countries[0] });
+        const remote = await client.redact(masked, {
+            country: countries[0], idempotencyKey: options.idempotencyKey,
+        });
         // A new array throughout: `local` may be the cached rules result.
         const [detections, exempted] = applyAllowlist(text, [...local.detections, ...ontoOriginal(remote.detections, masked, text, labels, CloudError)], allowed, domains);
         detections.sort((a, b) => a.start - b.start || b.end - a.end);
@@ -539,6 +544,7 @@ export class EuRedact {
             source: "cloud",
             tokens: tokenMapper ? tokenMapper.tokens : {},
             exempted,
+            ...(remote.cloud ? { cloud: remote.cloud } : {}),
         };
     }
     redact(text, options = {}) {
@@ -560,6 +566,9 @@ export class EuRedact {
         if (requestedMode !== "rules") {
             throw new Error(`unknown mode ${JSON.stringify(requestedMode)}: expected "rules" or "cloud"`);
         }
+        if (options.idempotencyKey !== undefined) {
+            throw new Error('idempotencyKey applies to mode: "cloud" only: a rules-only redaction makes no request to deduplicate');
+        }
         return this.redactRules(text, options, allowed, domains);
     }
     /** The local pipeline: normalise, detect, exempt, mask. */

package/dist/esm/types.d.ts

@@ -69,7 +69,8 @@ export declare enum EntityType {
     BIOMETRIC_REF = "BIOMETRIC_REF",
     FINANCIAL_AMOUNT = "FINANCIAL_AMOUNT",
     QUASI_IDENTIFIER = "QUASI_IDENTIFIER",
-    CREDENTIAL = "CREDENTIAL",
+    /** Legacy alias of SECRET: every credential, the model's included, is SECRET (rules-engine#111). */
+    CREDENTIAL = "SECRET",
     URL = "URL",
     OTHER = "OTHER"
 }
@@ -116,7 +117,13 @@ export interface Detection {
     text: string;
     source: DetectionSource;
     country: string | null;
-    confidence: string;
+    /**
+     * How the type was arrived at: "high", "medium" or "low" for a rules
+     * detection. `null` for a span the cloud model found and reported no
+     * confidence for: the model emits none today, and "high" would claim the
+     * checksum-backed evidence a rules detection has (rules-engine#108).
+     */
+    confidence: string | null;
     /**
      * How strongly the document supports {@link Detection.country}, in [0, 1].
      *
@@ -151,6 +158,52 @@ export interface Exemption {
     /** `"value"` for the literal allowlist, `"domain"` for allowlistDomains. */
     ruleKind: "value" | "domain";
 }
+/**
+ * One reason a cloud request cost what it did. `code` is one of the service's
+ * factor codes (`rules_only`, `document_length`, `instructions`,
+ * `dense_document`, `special_category_check`, `long_document`, `batch_rate`),
+ * kept as a string so a code added later on the service still arrives.
+ */
+export interface UsageFactor {
+    code: string;
+    /** One sentence explaining the factor, written by the service. */
+    detail: string;
+    /** `special_category_check` only: the kinds of data that triggered it. */
+    types?: string[];
+}
+/**
+ * What a cloud request cost, and why (rules-engine#89). A total and its
+ * reasons, never a cost per step; `credits` is `round(tokens * billingRate)`,
+ * the formula the service debits by.
+ */
+export interface Usage {
+    tokens: number;
+    billingRate: number;
+    credits: number;
+    factors: UsageFactor[];
+}
+/**
+ * Something the model found that could not be placed in the document: the
+ * service could not match it to a position in what it was sent, so nothing was
+ * masked for it. Reported so a caller can decide, rather than never being told.
+ */
+export interface Unlocated {
+    text: string;
+    entityType: EntityType | string;
+}
+/** What the cloud service said about a request, beside its spans (rules-engine#89). */
+export interface CloudInfo {
+    /** The service's job id: for support, for `Jobs.retrieve`, and to recognise an
+     *  idempotent replay. `null` for a batch document. */
+    jobId: string | null;
+    /** The public name of the model that answered, `null` when only rules ran. */
+    modelVersion: string | null;
+    /** What the request cost and why. Absent when the service does not report it
+     *  (older versions, batch documents) or reported it malformed. */
+    usage?: Usage;
+    /** What the model found but could not place in the text it was sent. */
+    unlocated: Unlocated[];
+}
 export interface RedactResult {
     redactedText: string;
     detections: Detection[];
@@ -188,6 +241,9 @@ export interface RedactResult {
     /** Detections the allowlist kept out of the output, with the rule that
      *  matched. Empty unless an allowlist was in force. */
     exempted: Exemption[];
+    /** What the cloud service reported beside the spans: job id, model, cost and
+     *  anything it could not place. Absent on a rules-only result. */
+    cloud?: CloudInfo;
 }
 export interface PatternDef {
     entityType: EntityType | string;

package/dist/esm/types.js

@@ -75,7 +75,8 @@ export var EntityType;
     EntityType["BIOMETRIC_REF"] = "BIOMETRIC_REF";
     EntityType["FINANCIAL_AMOUNT"] = "FINANCIAL_AMOUNT";
     EntityType["QUASI_IDENTIFIER"] = "QUASI_IDENTIFIER";
-    EntityType["CREDENTIAL"] = "CREDENTIAL";
+    /** Legacy alias of SECRET: every credential, the model's included, is SECRET (rules-engine#111). */
+    EntityType["CREDENTIAL"] = "SECRET";
     EntityType["URL"] = "URL";
     EntityType["OTHER"] = "OTHER";
 })(EntityType || (EntityType = {}));
@@ -89,6 +90,8 @@ export const LEGACY_TYPE_ALIASES = {
     NAME: "PERSON_NAME",
     STREET_ADDRESS: "ADDRESS",
     NATIONALITY_ETHNICITY: "SENSITIVE_ATTRIBUTE",
+    // The model labels prose credentials CREDENTIAL; the SDK reports every credential as SECRET (rules-engine#111).
+    CREDENTIAL: "SECRET",
 };
 /**
  * Resolve a possibly-legacy type name to its canonical form.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "euredact",
-  "version": "0.6.0",
+  "version": "0.7.0",
   "description": "European PII redaction SDK -- fast, zero-dependency, rule-based detection and redaction of personal data across 31 European countries",
   "homepage": "https://euredact.dev",
   "license": "Apache-2.0",
@@ -20,10 +20,15 @@
       }
     }
   },
+  "browser": {
+    "./dist/esm/platform.js": "./dist/esm/platform.browser.js",
+    "./dist/cjs/platform.js": "./dist/cjs/platform.browser.js"
+  },
   "files": [
     "dist",
     "README.md",
-    "LICENSE"
+    "LICENSE",
+    "NOTICE"
   ],
   "engines": {
     "node": ">=16.0.0"
@@ -54,7 +59,7 @@
     "build:cjs": "tsc -p tsconfig.cjs.json",
     "postbuild:cjs": "echo '{\"type\":\"commonjs\"}' > dist/cjs/package.json",
     "build": "rm -rf dist && npm run build:esm && npm run build:cjs",
-    "test": "npm run test:spec && npm run test:inference && npm run test:cloud && npm run test:tokenize && npm run test:allowlist && npm run test:eval-scoring && npm run test:conformance",
+    "test": "npm run test:spec && npm run test:inference && npm run test:cloud && npm run test:cloud-api && npm run test:tokenize && npm run test:allowlist && npm run test:eval-scoring && npm run test:platform && npm run test:batches && npm run test:conformance",
     "benchmark": "node --import tsx src/__tests__/benchmark.ts",
     "prepublishOnly": "npm run build",
     "test:spec": "node --import tsx src/__tests__/spec.ts",
@@ -64,9 +69,12 @@
     "test:inference": "node --import tsx src/__tests__/inference.ts",
     "dump": "node --import tsx src/__tests__/dump.ts",
     "test:cloud": "node --import tsx src/__tests__/cloud.ts",
+    "test:cloud-api": "node --import tsx src/__tests__/cloudApi.ts",
     "test:tokenize": "node --import tsx src/__tests__/tokenize.ts",
     "test:allowlist": "node --import tsx src/__tests__/allowlist.ts",
-    "test:eval-scoring": "node --import tsx src/__tests__/evalScoring.ts"
+    "test:eval-scoring": "node --import tsx src/__tests__/evalScoring.ts",
+    "test:platform": "node --import tsx src/__tests__/platform.ts",
+    "test:batches": "node --import tsx src/__tests__/batches.ts"
   },
   "devDependencies": {
     "@types/node": "22.10.2",