@ixo/decision-oracle 0.0.0-stage → 0.2.0

package/dist/lib/facts/stock/normalize.js

@@ -0,0 +1,135 @@
+/**
+ * §4.1 canonicalization — the anti-sneak pipeline behind the duplicate-detection body digest
+ * (duplicate-claim spec §4 L1). Applied per LEAF value; `canonicalBody` walks the claim body.
+ *
+ * Pure module: no I/O, no clock, no bindings — it runs inside the resolve phase (unique.ts) but everything
+ * here is deterministic compute, so review-replay re-derives identical output from the same body.
+ *
+ * Pipeline order is normative (spec §4.1): 1 NFKC → 2 casefold → 3 strip Cf → 4 confusable fold
+ * → 5 collapse whitespace → 6 strip separators → 7 date/number canonical forms. Steps 6–7
+ * interlock: the date and number branches read the PRE-strip form, because step 6 destroys
+ * exactly the structure (hyphens, decimal dot) those branches canonicalize — a value that is
+ * neither gets the plain separator strip. `test/unique-normalize.test.ts` pins the equivalences.
+ */
+/**
+ * §4.1 step 4 — pinned TR39-style confusable tables. VERSIONED DATA: any edit shifts every
+ * registered digest, so it is a format change — bump the version, expect re-registration.
+ * Keys are the CASE-FOLDED forms (the fold runs after step 2), hence lowercase Cyrillic/Greek.
+ */
+export const CONFUSABLES_VERSION = 1;
+const SCRIPT_FOLD = {
+    // Cyrillic → Latin lookalikes (keys look Latin — they are not; code points noted)
+    а: "a", // U+0430
+    е: "e", // U+0435
+    о: "o", // U+043E
+    і: "i", // U+0456 (Byelorussian-Ukrainian)
+    р: "p", // U+0440
+    с: "c", // U+0441
+    х: "x", // U+0445
+    у: "y", // U+0443
+    // Greek → Latin lookalikes (case-folded forms of Α Β Ε Ο Ρ Τ)
+    α: "a", // U+03B1
+    β: "b", // U+03B2
+    ε: "e", // U+03B5
+    ο: "o", // U+03BF
+    ρ: "p", // U+03C1
+    τ: "t", // U+03C4
+};
+/** Letter→digit fold, applied AFTER the script fold so e.g. Cyrillic о lands on 0 via o. */
+const DIGIT_FOLD = { o: "0", l: "1", i: "1" };
+/** RFC3339-ish, matched against the case-folded string (lowercase t/z; space works too). */
+const ISO_DATE_RE = /^(\d{4})-(\d{2})-(\d{2})(?:[t ]\d{2}:\d{2}(?::\d{2}(?:\.\d+)?)?(?:z|[+-]\d{2}:?\d{2})?)?$/;
+/** §4.1 step 6 — separator punctuation. */
+const SEPARATOR_RE = /[-_/.,#]/g;
+/** Step 7's number probe: grouping separators only — the decimal dot must survive the test. */
+const GROUPING_RE = /[-_/,#]/g;
+function foldConfusables(s) {
+    let out = "";
+    for (const ch of s) {
+        const latin = SCRIPT_FOLD[ch] ?? ch;
+        out += DIGIT_FOLD[latin] ?? latin;
+    }
+    return out;
+}
+/** Trim leading integer zeros + trailing fraction zeros; drop an emptied fraction outright. */
+function canonicalDecimal(s) {
+    const [int = "", frac = ""] = s.split(".");
+    const lead = int.replace(/^0+(?=\d)/, "");
+    const tail = frac.replace(/0+$/, "");
+    return tail ? `${lead}.${tail}` : lead;
+}
+/**
+ * Normalize one leaf value per §4.1. `granularity` only applies when the value IS a date —
+ * it comes from the rule's `by` entry (`{ ref, granularity }`), whole-body mode never sets it.
+ */
+export function normalizeScalar(v, opts) {
+    if (v === null || v === undefined)
+        return "";
+    // Steps 1–5. `toLowerCase` after NFKC is an acceptable casefold for the scripts we fold.
+    const folded = foldConfusables(String(v)
+        .normalize("NFKC")
+        .toLowerCase()
+        .replace(/\p{Cf}/gu, ""))
+        .replace(/\s+/gu, " ")
+        .trim();
+    // Step 7, dates — before the separator strip (it would eat the hyphens the form is made of).
+    const date = ISO_DATE_RE.exec(folded);
+    if (date) {
+        const [, y, mo, d] = date;
+        const moN = Number(mo);
+        const dN = Number(d);
+        if (moN >= 1 && moN <= 12 && dN >= 1 && dN <= 31) {
+            const g = opts?.granularity ?? "day";
+            return g === "year" ? y : g === "month" ? `${y}-${mo}` : `${y}-${mo}-${d}`;
+        }
+    }
+    // Step 7, numbers — grouping separators stripped, decimal dot kept for the probe.
+    const grouped = folded.replace(GROUPING_RE, "");
+    if (/^\d+(\.\d+)?$/.test(grouped))
+        return canonicalDecimal(grouped);
+    // Step 6 for everything else.
+    return folded.replace(SEPARATOR_RE, "");
+}
+/**
+ * Canonicalize a pulled claim body (spec §3 step 1): drop the top-level `credentials` block
+ * (VC/proof noise — never part of what was claimed), deep-sort object keys, normalize every
+ * leaf. Arrays keep their order — position is meaning in SurveyJS answers.
+ *
+ * `normalizedText` is every normalized leaf joined with "" in traversal order — the trigram
+ * lane's input (L3), erasable PII where the digest is not.
+ */
+export function canonicalBody(body) {
+    const leaves = [];
+    const walk = (v) => {
+        if (Array.isArray(v))
+            return v.map(walk);
+        if (v !== null && typeof v === "object") {
+            const out = {};
+            for (const k of Object.keys(v).sort())
+                out[k] = walk(v[k]);
+            return out;
+        }
+        const s = normalizeScalar(v);
+        leaves.push(s);
+        return s;
+    };
+    const canonical = {};
+    for (const k of Object.keys(body).sort()) {
+        if (k === "credentials")
+            continue;
+        canonical[k] = walk(body[k]);
+    }
+    return { canonical, normalizedText: leaves.join("") };
+}
+/**
+ * Keyed digest for the registry (spec §4 L1): HMAC-SHA256 hex over `parts.join("")`. Keyed
+ * because answers are low-entropy PII — a plain hash is dictionary-reversible; the keyed digest
+ * is what lets `claim_body_digests` outlive lawful erasure of `normalizedText`.
+ */
+export async function hmacDigest(key, parts) {
+    const enc = new TextEncoder();
+    const cryptoKey = await crypto.subtle.importKey("raw", enc.encode(key), { name: "HMAC", hash: "SHA-256" }, false, ["sign"]);
+    const sig = await crypto.subtle.sign("HMAC", cryptoKey, enc.encode(parts.join("")));
+    return [...new Uint8Array(sig)].map((b) => b.toString(16).padStart(2, "0")).join("");
+}
+//# sourceMappingURL=normalize.js.map

package/dist/lib/facts/stock/normalize.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"normalize.js","sourceRoot":"","sources":["../../../../src/lib/facts/stock/normalize.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH;;;;GAIG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAErC,MAAM,WAAW,GAA2B;IAC1C,kFAAkF;IAClF,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,kCAAkC;IAC1C,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,8DAA8D;IAC9D,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;IACjB,CAAC,EAAE,GAAG,EAAE,SAAS;CAClB,CAAC;AAEF,4FAA4F;AAC5F,MAAM,UAAU,GAA2B,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC;AAEtE,4FAA4F;AAC5F,MAAM,WAAW,GACf,2FAA2F,CAAC;AAE9F,2CAA2C;AAC3C,MAAM,YAAY,GAAG,WAAW,CAAC;AACjC,+FAA+F;AAC/F,MAAM,WAAW,GAAG,UAAU,CAAC;AAE/B,SAAS,eAAe,CAAC,CAAS;IAChC,IAAI,GAAG,GAAG,EAAE,CAAC;IACb,KAAK,MAAM,EAAE,IAAI,CAAC,EAAE,CAAC;QACnB,MAAM,KAAK,GAAG,WAAW,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;QACpC,GAAG,IAAI,UAAU,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC;IACpC,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,+FAA+F;AAC/F,SAAS,gBAAgB,CAAC,CAAS;IACjC,MAAM,CAAC,GAAG,GAAG,EAAE,EAAE,IAAI,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC3C,MAAM,IAAI,GAAG,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;IAC1C,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;IACrC,OAAO,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;AACzC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAC7B,CAAU,EACV,IAAiD;IAEjD,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,KAAK,SAAS;QAAE,OAAO,EAAE,CAAC;IAC7C,yFAAyF;IACzF,MAAM,MAAM,GAAG,eAAe,CAC5B,MAAM,CAAC,CAAC,CAAC;SACN,SAAS,CAAC,MAAM,CAAC;SACjB,WAAW,EAAE;SACb,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAC3B;SACE,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC;SACrB,IAAI,EAAE,CAAC;IAEV,6FAA6F;IAC7F,MAAM,IAAI,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACtC,IAAI,IAAI,EAAE,CAAC;QACT,MAAM,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,GAAG,IAAmD,CAAC;QACzE,MAAM,GAAG,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC;QACvB,MAAM,EAAE,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;QACrB,IAAI,GAAG,IAAI,CAAC,IAAI,GAAG,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC;YACjD,MAAM,CAAC,GAAG,IAAI,EAAE,WAAW,IAAI,KAAK,CAAC;YACrC,OAAO,CAAC,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,CAAC;QAC7E,CAAC;IACH,CAAC;IAED,kFAAkF;IAClF,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;IAChD,IAAI,eAAe,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,OAAO,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAEpE,8BAA8B;IAC9B,OAAO,MAAM,CAAC,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC;AAC1C,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,aAAa,CAAC,IAA6B;IAIzD,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,MAAM,IAAI,GAAG,CAAC,CAAU,EAAW,EAAE;QACnC,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;YAAE,OAAO,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACzC,IAAI,CAAC,KAAK,IAAI,IAAI,OAAO,CAAC,KAAK,QAAQ,EAAE,CAAC;YACxC,MAAM,GAAG,GAA4B,EAAE,CAAC;YACxC,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE;gBAAE,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,CAAE,CAA6B,CAAC,CAAC,CAAC,CAAC,CAAC;YACxF,OAAO,GAAG,CAAC;QACb,CAAC;QACD,MAAM,CAAC,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC;QAC7B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACf,OAAO,CAAC,CAAC;IACX,CAAC,CAAC;IAEF,MAAM,SAAS,GAA4B,EAAE,CAAC;IAC9C,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;QACzC,IAAI,CAAC,KAAK,aAAa;YAAE,SAAS;QAClC,SAAS,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/B,CAAC;IACD,OAAO,EAAE,SAAS,EAAE,cAAc,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC;AACxD,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,GAAW,EAAE,KAAe;IAC3D,MAAM,GAAG,GAAG,IAAI,WAAW,EAAE,CAAC;IAC9B,MAAM,SAAS,GAAG,MAAM,MAAM,CAAC,MAAM,CAAC,SAAS,CAC7C,KAAK,EACL,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,EACf,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,SAAS,EAAE,EACjC,KAAK,EACL,CAAC,MAAM,CAAC,CACT,CAAC;IACF,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE,SAAS,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IACpF,OAAO,CAAC,GAAG,IAAI,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AACvF,CAAC"}

package/dist/lib/facts/stock/phash.d.ts

@@ -0,0 +1,55 @@
+/**
+ * Perceptual hash (dHash64) for duplicate-attachment detection — spec §4 L4.
+ *
+ * Catches re-scanned/re-exported files whose bytes (and CID) changed but whose pixels didn't:
+ * grayscale → box-downsample to a 9×8 grid → 64 row-gradient bits (`g[y][x] > g[y][x+1]`) → 16
+ * hex chars. Gradients survive re-encode/rescale noise that raw pixels don't, which is the whole
+ * trick. Hamming distance between two hashes maps onto the §5 score tiers (≤6 / ≤10).
+ *
+ * PURE and deterministic — this runs in the resolve phase (unique.ts) and the resulting hash is
+ * frozen into the FactSet; review-replay never re-decodes. Decode failures return `null`, never
+ * throw: an undecodable attachment simply contributes no pHash and the text lanes still run.
+ * jpeg-js / upng-js are small pure-JS decoders (no wasm, no lazy-load ceremony needed).
+ */
+/** RGBA pixel buffer, 4 bytes per pixel. */
+export interface Pixels {
+    width: number;
+    height: number;
+    data: Uint8Array | Uint8ClampedArray;
+}
+/** hamming ≤ 10 ⇒ near-match (spec §5 maps ≤6 → 8500 bps, ≤10 → 7000 bps). */
+export declare const PHASH_NEAR = 10;
+/** What `pngDecodeFootprint` reads off the header, without inflating anything. */
+export interface PngHeader {
+    width: number;
+    height: number;
+    bitDepth: number;
+    colorType: number;
+    interlace: number;
+    /** The IDAT stream must inflate to exactly this many bytes: filter byte + scanline, per row. */
+    inflatedBytes: number;
+    /** Bytes UPNG holds at the peak of decoding this image — see `MAX_PNG_DECODE_BYTES`. */
+    footprint: number;
+}
+/**
+ * The PNG's IHDR plus what decoding it would cost, read by walking the chunk headers only. Null
+ * when the bytes are not a PNG, are truncated, carry no IHDR, declare an unknown colour type /
+ * bit depth / interlace method, have no pixels, or are an APNG — none of which this lane hashes.
+ */
+export declare function pngDecodeFootprint(bytes: Uint8Array): PngHeader | null;
+/**
+ * Inflate a zlib stream into a buffer of exactly `size` bytes; a stream that would run past it, or
+ * stops short of it, throws. The chunks land straight in the output (no chunk list + flatten copy,
+ * which is how pako and `fflate.unzlibSync` momentarily double the inflated size).
+ */
+export declare function inflateExactly(data: Uint8Array, size: number): Uint8Array;
+/** 16-hex-char row-gradient dHash of an RGBA buffer. Pure math — safe on any dimensions ≥ 1. */
+export declare function dhash64(p: Pixels): string;
+/** Bit distance between two 16-hex-char hashes (0–64). */
+export declare function hamming(a: string, b: string): number;
+/** Decode image bytes to RGBA. `image/jpeg` via jpeg-js, `image/png` via UPNG; anything else — or
+ *  malformed bytes — is `null` (resolve-phase callers record and move on, they never catch). */
+export declare function decodePixels(bytes: Uint8Array, mediaType: string): Pixels | null;
+/** decode → dhash64; `null` if the bytes are oversized (> 10 MB) or undecodable. */
+export declare function phashImage(bytes: Uint8Array, mediaType: string): string | null;
+//# sourceMappingURL=phash.d.ts.map

package/dist/lib/facts/stock/phash.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"phash.d.ts","sourceRoot":"","sources":["../../../../src/lib/facts/stock/phash.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAMH,4CAA4C;AAC5C,MAAM,WAAW,MAAM;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,UAAU,GAAG,iBAAiB,CAAC;CACtC;AAED,8EAA8E;AAC9E,eAAO,MAAM,UAAU,KAAK,CAAC;AAkD7B,kFAAkF;AAClF,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,gGAAgG;IAChG,aAAa,EAAE,MAAM,CAAC;IACtB,wFAAwF;IACxF,SAAS,EAAE,MAAM,CAAC;CACnB;AAQD;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,UAAU,GAAG,SAAS,GAAG,IAAI,CAoDtE;AASD;;;;GAIG;AACH,wBAAgB,cAAc,CAAC,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,GAAG,UAAU,CAczE;AAMD,gGAAgG;AAChG,wBAAgB,OAAO,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CA+BzC;AAED,0DAA0D;AAC1D,wBAAgB,OAAO,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAUpD;AAED;gGACgG;AAChG,wBAAgB,YAAY,CAAC,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAiChF;AAED,oFAAoF;AACpF,wBAAgB,UAAU,CAAC,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAI9E"}

package/dist/lib/facts/stock/phash.js

@@ -0,0 +1,256 @@
+/**
+ * Perceptual hash (dHash64) for duplicate-attachment detection — spec §4 L4.
+ *
+ * Catches re-scanned/re-exported files whose bytes (and CID) changed but whose pixels didn't:
+ * grayscale → box-downsample to a 9×8 grid → 64 row-gradient bits (`g[y][x] > g[y][x+1]`) → 16
+ * hex chars. Gradients survive re-encode/rescale noise that raw pixels don't, which is the whole
+ * trick. Hamming distance between two hashes maps onto the §5 score tiers (≤6 / ≤10).
+ *
+ * PURE and deterministic — this runs in the resolve phase (unique.ts) and the resulting hash is
+ * frozen into the FactSet; review-replay never re-decodes. Decode failures return `null`, never
+ * throw: an undecodable attachment simply contributes no pHash and the text lanes still run.
+ * jpeg-js / upng-js are small pure-JS decoders (no wasm, no lazy-load ceremony needed).
+ */
+import { Unzlib } from "fflate";
+import { decode as jpegDecode } from "jpeg-js";
+import UPNG from "upng-js";
+/** hamming ≤ 10 ⇒ near-match (spec §5 maps ≤6 → 8500 bps, ≤10 → 7000 bps). */
+export const PHASH_NEAR = 10;
+// Same cap as MAX_AI_ATTACHMENT_BYTES (unique.ts) — an attachment past the AI lanes' budget is
+// past ours too, and jpeg-js on adversarially large inputs is pure-CPU pain.
+const MAX_IMAGE_BYTES = 10 * 1024 * 1024;
+/**
+ * The most memory a PNG may need to DECODE. The byte cap above bounds the file, not the image:
+ * deflate squeezes a blank 8000×8000 canvas into a few hundred KB, and UPNG then holds the inflated
+ * scanlines AND the RGBA8 copy at once — hundreds of MB, past the isolate's 128 MB, which is not a
+ * catchable throw but the end of the step and of every claim sharing the isolate. So the header is
+ * read first and the decode is costed from it (`pngDecodeFootprint`): the scanline bytes the IDAT
+ * stream must inflate to (bit depth × channels, Adam7 passes included), the de-interlaced copy
+ * UPNG makes for an interlaced image, the 4-byte-per-pixel RGBA8 frame — and three file-sized
+ * buffers alive beside them (the caller's bytes, UPNG's zero-padded IDAT copy, and one inflate
+ * push's transient output or the exact-buffer copy `decodePixels` makes of a view), which matter
+ * once the file itself is megabytes. 64 MiB is the same budget the JPEG decoder is given below: it
+ * keeps an 8-bit RGB 4K screenshot (~55 MiB decoded) inside for a file up to ~3 MB, while an RGBA
+ * one (~63 MiB decoded) is past it. Anything larger simply has no pHash (`null`), which only
+ * narrows recall — never a failure.
+ *
+ * The footprint is honest only because UPNG's inflater is replaced (see `decodePixels`): 2.1.0
+ * inflates with `pako.inflate`, which has no output bound and whatever IHDR says — a 1×1 header
+ * with 10 MB of deflated zeros behind it would still inflate to gigabytes. The bounded inflater
+ * writes into a buffer of exactly the costed scanline size and aborts the moment the stream runs
+ * past it, so a lying IDAT is a caught throw, not an OOM.
+ */
+const MAX_PNG_DECODE_BYTES = 64 * 1024 * 1024;
+const PNG_SIGNATURE = [0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a];
+const IHDR = 0x49484452;
+const IEND = 0x49454e44;
+/** APNG chunks. UPNG decodes every frame at the frame's OWN declared size (never checked against
+ *  IHDR) and allocates an RGBA8 frame each — a hash of frame 0 is not worth that exposure. */
+const APNG_CHUNKS = new Set([0x6163544c /* acTL */, 0x6663544c /* fcTL */, 0x66644154 /* fdAT */]);
+/** Samples per pixel by PNG colour type (PNG spec §11.2.2); any other type is malformed. */
+const PNG_CHANNELS = { 0: 1, 2: 3, 3: 1, 4: 2, 6: 4 };
+const PNG_BIT_DEPTHS = new Set([1, 2, 4, 8, 16]);
+/** Adam7 passes as `[x0, y0, dx, dy]` (PNG spec §8.2) — a pass with no pixels has no scanlines. */
+const ADAM7 = [
+    [0, 0, 8, 8],
+    [4, 0, 8, 8],
+    [0, 4, 4, 8],
+    [2, 0, 4, 4],
+    [0, 2, 2, 4],
+    [1, 0, 2, 2],
+    [0, 1, 1, 2],
+];
+/** Filter byte + packed scanline, per row (PNG spec §7.2); a pass with no pixels has none. */
+function scanlineBytes(width, height, bitsPerPixel) {
+    if (width === 0 || height === 0)
+        return 0;
+    return height * (1 + Math.ceil((width * bitsPerPixel) / 8));
+}
+/**
+ * The PNG's IHDR plus what decoding it would cost, read by walking the chunk headers only. Null
+ * when the bytes are not a PNG, are truncated, carry no IHDR, declare an unknown colour type /
+ * bit depth / interlace method, have no pixels, or are an APNG — none of which this lane hashes.
+ */
+export function pngDecodeFootprint(bytes) {
+    if (bytes.byteLength < PNG_SIGNATURE.length)
+        return null;
+    if (!PNG_SIGNATURE.every((byte, i) => bytes[i] === byte))
+        return null;
+    const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength);
+    let header = null;
+    // Each chunk: length (u32, big-endian) · type (4 bytes) · data · CRC (4 bytes). Apple's `CgBI`
+    // precedes IHDR, so the walk does not insist on IHDR coming first.
+    let at = PNG_SIGNATURE.length;
+    while (at + 8 <= bytes.byteLength) {
+        const length = view.getUint32(at, false);
+        const type = view.getUint32(at + 4, false);
+        // A chunk whose data runs past the end is a truncated file — UPNG would read zeros for it.
+        if (at + 12 + length > bytes.byteLength)
+            return null;
+        if (APNG_CHUNKS.has(type))
+            return null;
+        if (type === IHDR) {
+            if (length < 13)
+                return null;
+            header = {
+                width: view.getUint32(at + 8, false),
+                height: view.getUint32(at + 12, false),
+                bitDepth: bytes[at + 16] ?? 0,
+                colorType: bytes[at + 17] ?? 0,
+                interlace: bytes[at + 20] ?? 0,
+            };
+        }
+        if (type === IEND)
+            break;
+        at += 12 + length;
+    }
+    if (!header || header.width === 0 || header.height === 0)
+        return null;
+    const channels = PNG_CHANNELS[header.colorType];
+    if (!channels || !PNG_BIT_DEPTHS.has(header.bitDepth))
+        return null;
+    const bitsPerPixel = channels * header.bitDepth;
+    let inflatedBytes;
+    if (header.interlace === 0) {
+        inflatedBytes = scanlineBytes(header.width, header.height, bitsPerPixel);
+    }
+    else if (header.interlace === 1) {
+        inflatedBytes = 0;
+        for (const [x0, y0, dx, dy] of ADAM7) {
+            const cols = header.width > x0 ? Math.ceil((header.width - x0) / dx) : 0;
+            const rows = header.height > y0 ? Math.ceil((header.height - y0) / dy) : 0;
+            inflatedBytes += scanlineBytes(cols, rows, bitsPerPixel);
+        }
+    }
+    else {
+        return null;
+    }
+    // The inflated scanlines (the bounded inflater's one buffer), the de-interlaced copy UPNG's
+    // `_readInterlace` allocates beside them, the RGBA8 frame `toRGBA8` builds, and the three
+    // file-sized buffers (see `MAX_PNG_DECODE_BYTES`).
+    const footprint = inflatedBytes * (header.interlace === 1 ? 2 : 1) +
+        header.width * header.height * 4 +
+        3 * bytes.byteLength;
+    return { ...header, inflatedBytes, footprint };
+}
+/**
+ * How much deflate input one `Unzlib.push` is handed. fflate inflates everything pushed before it
+ * reports any output, and deflate expands at most ~1032:1, so 4 KiB in bounds one push's output at
+ * ~4 MiB — the most a lying stream can allocate before the size check aborts it.
+ */
+const INFLATE_PUSH_BYTES = 4096;
+/**
+ * Inflate a zlib stream into a buffer of exactly `size` bytes; a stream that would run past it, or
+ * stops short of it, throws. The chunks land straight in the output (no chunk list + flatten copy,
+ * which is how pako and `fflate.unzlibSync` momentarily double the inflated size).
+ */
+export function inflateExactly(data, size) {
+    const out = new Uint8Array(size);
+    let at = 0;
+    const unzlib = new Unzlib((chunk) => {
+        if (at + chunk.length > size)
+            throw new Error("zlib stream inflates past its declared size");
+        out.set(chunk, at);
+        at += chunk.length;
+    });
+    for (let off = 0; off < data.length; off += INFLATE_PUSH_BYTES) {
+        const end = Math.min(off + INFLATE_PUSH_BYTES, data.length);
+        unzlib.push(data.subarray(off, end), end === data.length);
+    }
+    if (at !== size)
+        throw new Error("zlib stream inflates short of its declared size");
+    return out;
+}
+// dHash grid: 9 columns × 8 rows of grayscale means ⇒ 8 gradient bits per row, 64 total.
+const GRID_W = 9;
+const GRID_H = 8;
+/** 16-hex-char row-gradient dHash of an RGBA buffer. Pure math — safe on any dimensions ≥ 1. */
+export function dhash64(p) {
+    // Box-downsample straight from RGBA to the grayscale grid — no intermediate full-size pass.
+    const grid = new Float64Array(GRID_W * GRID_H);
+    for (let gy = 0; gy < GRID_H; gy++) {
+        const y0 = Math.floor((gy * p.height) / GRID_H);
+        const y1 = Math.max(y0 + 1, Math.floor(((gy + 1) * p.height) / GRID_H));
+        for (let gx = 0; gx < GRID_W; gx++) {
+            const x0 = Math.floor((gx * p.width) / GRID_W);
+            const x1 = Math.max(x0 + 1, Math.floor(((gx + 1) * p.width) / GRID_W));
+            let sum = 0;
+            for (let y = y0; y < y1; y++) {
+                for (let x = x0; x < x1; x++) {
+                    const i = (y * p.width + x) * 4;
+                    // ITU-R BT.601 luma weights.
+                    sum +=
+                        0.299 * (p.data[i] ?? 0) + 0.587 * (p.data[i + 1] ?? 0) + 0.114 * (p.data[i + 2] ?? 0);
+                }
+            }
+            grid[gy * GRID_W + gx] = sum / ((y1 - y0) * (x1 - x0));
+        }
+    }
+    let hex = "";
+    for (let gy = 0; gy < GRID_H; gy++) {
+        let byte = 0;
+        for (let gx = 0; gx < GRID_W - 1; gx++) {
+            const bit = (grid[gy * GRID_W + gx] ?? 0) > (grid[gy * GRID_W + gx + 1] ?? 0) ? 1 : 0;
+            byte = (byte << 1) | bit;
+        }
+        hex += byte.toString(16).padStart(2, "0");
+    }
+    return hex;
+}
+/** Bit distance between two 16-hex-char hashes (0–64). */
+export function hamming(a, b) {
+    let dist = 0;
+    for (let i = 0; i < 16; i++) {
+        let x = parseInt(a[i] ?? "0", 16) ^ parseInt(b[i] ?? "0", 16);
+        while (x) {
+            dist += x & 1;
+            x >>= 1;
+        }
+    }
+    return dist;
+}
+/** Decode image bytes to RGBA. `image/jpeg` via jpeg-js, `image/png` via UPNG; anything else — or
+ *  malformed bytes — is `null` (resolve-phase callers record and move on, they never catch). */
+export function decodePixels(bytes, mediaType) {
+    const mime = (mediaType.split(";")[0] ?? "").trim().toLowerCase();
+    try {
+        if (mime === "image/jpeg") {
+            const img = jpegDecode(bytes, { useTArray: true, maxMemoryUsageInMB: 64 });
+            return { width: img.width, height: img.height, data: img.data };
+        }
+        if (mime === "image/png") {
+            // Refuse a decompression bomb BEFORE inflating it (see `MAX_PNG_DECODE_BYTES`): a header
+            // that is missing, or whose decode would cost more than the budget, is an image this lane
+            // does not hash.
+            const header = pngDecodeFootprint(bytes);
+            if (header === null || header.footprint > MAX_PNG_DECODE_BYTES)
+                return null;
+            // A Uint8Array view may span a larger buffer — hand UPNG an exact copy (c2pa.ts precedent),
+            // but only then: a fetched attachment already owns its whole buffer.
+            const buf = (bytes.byteOffset === 0 && bytes.byteLength === bytes.buffer.byteLength
+                ? bytes.buffer
+                : bytes.buffer.slice(bytes.byteOffset, bytes.byteOffset + bytes.byteLength));
+            // UPNG hands its inflater the whole file-sized IDAT buffer (zero-padded past the stream);
+            // the bounded one stops at the costed scanline size whatever the stream claims. Rebound on
+            // every decode so the budget it enforces is visibly this image's.
+            UPNG.decode._inflate = (data) => inflateExactly(data, header.inflatedBytes);
+            const img = UPNG.decode(buf);
+            const frame = UPNG.toRGBA8(img)[0];
+            if (!frame)
+                return null;
+            return { width: img.width, height: img.height, data: new Uint8Array(frame) };
+        }
+    }
+    catch {
+        // fall through — undecodable is a recorded non-fact, not an error
+    }
+    return null;
+}
+/** decode → dhash64; `null` if the bytes are oversized (> 10 MB) or undecodable. */
+export function phashImage(bytes, mediaType) {
+    if (bytes.byteLength > MAX_IMAGE_BYTES)
+        return null;
+    const pixels = decodePixels(bytes, mediaType);
+    return pixels ? dhash64(pixels) : null;
+}
+//# sourceMappingURL=phash.js.map

package/dist/lib/facts/stock/phash.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"phash.js","sourceRoot":"","sources":["../../../../src/lib/facts/stock/phash.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;AAChC,OAAO,EAAE,MAAM,IAAI,UAAU,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,IAAI,MAAM,SAAS,CAAC;AAS3B,8EAA8E;AAC9E,MAAM,CAAC,MAAM,UAAU,GAAG,EAAE,CAAC;AAE7B,+FAA+F;AAC/F,6EAA6E;AAC7E,MAAM,eAAe,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AAEzC;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,oBAAoB,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AAE9C,MAAM,aAAa,GAAG,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;AACvE,MAAM,IAAI,GAAG,UAAU,CAAC;AACxB,MAAM,IAAI,GAAG,UAAU,CAAC;AACxB;8FAC8F;AAC9F,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,CAAC,UAAU,CAAC,UAAU,EAAE,UAAU,CAAC,UAAU,EAAE,UAAU,CAAC,UAAU,CAAC,CAAC,CAAC;AAEnG,4FAA4F;AAC5F,MAAM,YAAY,GAA2B,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC;AAC9E,MAAM,cAAc,GAAG,IAAI,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;AACjD,mGAAmG;AACnG,MAAM,KAAK,GAAG;IACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;CACJ,CAAC;AAeX,8FAA8F;AAC9F,SAAS,aAAa,CAAC,KAAa,EAAE,MAAc,EAAE,YAAoB;IACxE,IAAI,KAAK,KAAK,CAAC,IAAI,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAC1C,OAAO,MAAM,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;AAC9D,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,kBAAkB,CAAC,KAAiB;IAClD,IAAI,KAAK,CAAC,UAAU,GAAG,aAAa,CAAC,MAAM;QAAE,OAAO,IAAI,CAAC;IACzD,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC;IACtE,MAAM,IAAI,GAAG,IAAI,QAAQ,CAAC,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,UAAU,EAAE,KAAK,CAAC,UAAU,CAAC,CAAC;IAC5E,IAAI,MAAM,GAA0D,IAAI,CAAC;IACzE,+FAA+F;IAC/F,mEAAmE;IACnE,IAAI,EAAE,GAAG,aAAa,CAAC,MAAM,CAAC;IAC9B,OAAO,EAAE,GAAG,CAAC,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;QAClC,MAAM,MAAM,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,CAAC,EAAE,KAAK,CAAC,CAAC;QAC3C,2FAA2F;QAC3F,IAAI,EAAE,GAAG,EAAE,GAAG,MAAM,GAAG,KAAK,CAAC,UAAU;YAAE,OAAO,IAAI,CAAC;QACrD,IAAI,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,OAAO,IAAI,CAAC;QACvC,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;YAClB,IAAI,MAAM,GAAG,EAAE;gBAAE,OAAO,IAAI,CAAC;YAC7B,MAAM,GAAG;gBACP,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,CAAC,EAAE,KAAK,CAAC;gBACpC,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,EAAE,KAAK,CAAC;gBACtC,QAAQ,EAAE,KAAK,CAAC,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC;gBAC7B,SAAS,EAAE,KAAK,CAAC,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC;gBAC9B,SAAS,EAAE,KAAK,CAAC,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC;aAC/B,CAAC;QACJ,CAAC;QACD,IAAI,IAAI,KAAK,IAAI;YAAE,MAAM;QACzB,EAAE,IAAI,EAAE,GAAG,MAAM,CAAC;IACpB,CAAC;IACD,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,KAAK,CAAC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACtE,MAAM,QAAQ,GAAG,YAAY,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;IAChD,IAAI,CAAC,QAAQ,IAAI,CAAC,cAAc,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC;QAAE,OAAO,IAAI,CAAC;IACnE,MAAM,YAAY,GAAG,QAAQ,GAAG,MAAM,CAAC,QAAQ,CAAC;IAChD,IAAI,aAAqB,CAAC;IAC1B,IAAI,MAAM,CAAC,SAAS,KAAK,CAAC,EAAE,CAAC;QAC3B,aAAa,GAAG,aAAa,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IAC3E,CAAC;SAAM,IAAI,MAAM,CAAC,SAAS,KAAK,CAAC,EAAE,CAAC;QAClC,aAAa,GAAG,CAAC,CAAC;QAClB,KAAK,MAAM,CAAC,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC,IAAI,KAAK,EAAE,CAAC;YACrC,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,GAAG,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACzE,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3E,aAAa,IAAI,aAAa,CAAC,IAAI,EAAE,IAAI,EAAE,YAAY,CAAC,CAAC;QAC3D,CAAC;IACH,CAAC;SAAM,CAAC;QACN,OAAO,IAAI,CAAC;IACd,CAAC;IACD,4FAA4F;IAC5F,0FAA0F;IAC1F,mDAAmD;IACnD,MAAM,SAAS,GACb,aAAa,GAAG,CAAC,MAAM,CAAC,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAChD,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC;QAChC,CAAC,GAAG,KAAK,CAAC,UAAU,CAAC;IACvB,OAAO,EAAE,GAAG,MAAM,EAAE,aAAa,EAAE,SAAS,EAAE,CAAC;AACjD,CAAC;AAED;;;;GAIG;AACH,MAAM,kBAAkB,GAAG,IAAI,CAAC;AAEhC;;;;GAIG;AACH,MAAM,UAAU,cAAc,CAAC,IAAgB,EAAE,IAAY;IAC3D,MAAM,GAAG,GAAG,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC;IACjC,IAAI,EAAE,GAAG,CAAC,CAAC;IACX,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE;QAClC,IAAI,EAAE,GAAG,KAAK,CAAC,MAAM,GAAG,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAC7F,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QACnB,EAAE,IAAI,KAAK,CAAC,MAAM,CAAC;IACrB,CAAC,CAAC,CAAC;IACH,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,IAAI,kBAAkB,EAAE,CAAC;QAC/D,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;QAC5D,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,GAAG,KAAK,IAAI,CAAC,MAAM,CAAC,CAAC;IAC5D,CAAC;IACD,IAAI,EAAE,KAAK,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,iDAAiD,CAAC,CAAC;IACpF,OAAO,GAAG,CAAC;AACb,CAAC;AAED,yFAAyF;AACzF,MAAM,MAAM,GAAG,CAAC,CAAC;AACjB,MAAM,MAAM,GAAG,CAAC,CAAC;AAEjB,gGAAgG;AAChG,MAAM,UAAU,OAAO,CAAC,CAAS;IAC/B,4FAA4F;IAC5F,MAAM,IAAI,GAAG,IAAI,YAAY,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC;IAC/C,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,MAAM,EAAE,EAAE,EAAE,EAAE,CAAC;QACnC,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC;QAChD,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC;QACxE,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,MAAM,EAAE,EAAE,EAAE,EAAE,CAAC;YACnC,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,KAAK,CAAC,GAAG,MAAM,CAAC,CAAC;YAC/C,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC;YACvE,IAAI,GAAG,GAAG,CAAC,CAAC;YACZ,KAAK,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC7B,KAAK,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;oBAC7B,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;oBAChC,6BAA6B;oBAC7B,GAAG;wBACD,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;gBAC3F,CAAC;YACH,CAAC;YACD,IAAI,CAAC,EAAE,GAAG,MAAM,GAAG,EAAE,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC;QACzD,CAAC;IACH,CAAC;IACD,IAAI,GAAG,GAAG,EAAE,CAAC;IACb,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,MAAM,EAAE,EAAE,EAAE,EAAE,CAAC;QACnC,IAAI,IAAI,GAAG,CAAC,CAAC;QACb,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,MAAM,GAAG,CAAC,EAAE,EAAE,EAAE,EAAE,CAAC;YACvC,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,EAAE,GAAG,MAAM,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,GAAG,MAAM,GAAG,EAAE,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACtF,IAAI,GAAG,CAAC,IAAI,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC;QAC3B,CAAC;QACD,GAAG,IAAI,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC5C,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,0DAA0D;AAC1D,MAAM,UAAU,OAAO,CAAC,CAAS,EAAE,CAAS;IAC1C,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5B,IAAI,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAE,EAAE,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAE,EAAE,CAAC,CAAC;QAC9D,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC;YACd,CAAC,KAAK,CAAC,CAAC;QACV,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED;gGACgG;AAChG,MAAM,UAAU,YAAY,CAAC,KAAiB,EAAE,SAAiB;IAC/D,MAAM,IAAI,GAAG,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAClE,IAAI,CAAC;QACH,IAAI,IAAI,KAAK,YAAY,EAAE,CAAC;YAC1B,MAAM,GAAG,GAAG,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,kBAAkB,EAAE,EAAE,EAAE,CAAC,CAAC;YAC3E,OAAO,EAAE,KAAK,EAAE,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,CAAC;QAClE,CAAC;QACD,IAAI,IAAI,KAAK,WAAW,EAAE,CAAC;YACzB,yFAAyF;YACzF,0FAA0F;YAC1F,iBAAiB;YACjB,MAAM,MAAM,GAAG,kBAAkB,CAAC,KAAK,CAAC,CAAC;YACzC,IAAI,MAAM,KAAK,IAAI,IAAI,MAAM,CAAC,SAAS,GAAG,oBAAoB;gBAAE,OAAO,IAAI,CAAC;YAC5E,4FAA4F;YAC5F,qEAAqE;YACrE,MAAM,GAAG,GAAG,CACV,KAAK,CAAC,UAAU,KAAK,CAAC,IAAI,KAAK,CAAC,UAAU,KAAK,KAAK,CAAC,MAAM,CAAC,UAAU;gBACpE,CAAC,CAAC,KAAK,CAAC,MAAM;gBACd,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,EAAE,KAAK,CAAC,UAAU,GAAG,KAAK,CAAC,UAAU,CAAC,CAC/D,CAAC;YACjB,0FAA0F;YAC1F,2FAA2F;YAC3F,kEAAkE;YAClE,IAAI,CAAC,MAAM,CAAC,QAAQ,GAAG,CAAC,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,MAAM,CAAC,aAAa,CAAC,CAAC;YAC5E,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7B,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;YACnC,IAAI,CAAC,KAAK;gBAAE,OAAO,IAAI,CAAC;YACxB,OAAO,EAAE,KAAK,EAAE,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,IAAI,EAAE,IAAI,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QAC/E,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,kEAAkE;IACpE,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,UAAU,CAAC,KAAiB,EAAE,SAAiB;IAC7D,IAAI,KAAK,CAAC,UAAU,GAAG,eAAe;QAAE,OAAO,IAAI,CAAC;IACpD,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC9C,OAAO,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;AACzC,CAAC"}

package/dist/lib/facts/stock/rule-defaults.d.ts

@@ -0,0 +1,72 @@
+/**
+ * The duplicate/frequency rules as the STOCK producer runs them — the engine's manifest
+ * declaration (lib/engine/types.ts `ManifestUnique` / `ManifestFrequency`, every parameter
+ * optional on the wire) resolved to a complete rule.
+ *
+ * THE defaults below mirror eval-engine `src/lib/rubric/schema.ts` `Unique` / `Frequency`
+ * (`countPrior`, `attachments`, `ai`, `thresholds`, `onBreach` — what `Unique.parse` /
+ * `Frequency.parse` yield when the author left a field out) and must be kept in LOCK-STEP with
+ * them until every engine sends the full post-default declaration (decision-engine @ 444aeae
+ * already does; an older engine sends `{ ruleId }` only). One place, on purpose: a default that
+ * drifted from the engine's would score a claim against numbers the rubric's author never set.
+ *
+ * `frequency` is different: `max` and `per` HAVE no default — a rule without them cannot run, so
+ * `frequencyRuleOf` answers null and the stock producer produces nothing; validate.ts then
+ * refuses the build (`FACT_MISSING /facts/frequency "manifest lacks max/per"` ⇒ `invalid_facts`,
+ * our/the engine's bug, never the claim's).
+ */
+import type { ManifestFrequency, ManifestUnique } from "../../engine/types.js";
+/** One entry of the identity key (`by`): a plain `$ref`, or a ref with a date granularity (dup spec §6). */
+export type UniqueBy = string | {
+    ref: string;
+    granularity?: "day" | "month" | "year";
+};
+export interface UniqueRule {
+    id: string;
+    /** Identity key — enables the evasion-asymmetry detector. Absent ⇒ whole canonical body. */
+    by?: UniqueBy[];
+    /** Which prior verdicts block — rejected priors never do (the honest-fix path, dup §6). */
+    countPrior: string[];
+    /** The attachment lanes: CID exact match + pHash near match (dup §4 L2/L4). */
+    attachments: boolean;
+    /** The gray-zone same-event AI judge (dup §7), bounded by maxPairs. */
+    ai: {
+        enabled: boolean;
+        maxPairs: number;
+    };
+    /** Band cutoffs (dup §5): score ≥ rejectAtBps ⇒ reject, ≥ reviewAtBps ⇒ review, else pass. */
+    thresholds: {
+        rejectAtBps: number;
+        reviewAtBps: number;
+    };
+}
+export type FrequencyPeriod = "day" | "week" | "month" | "year";
+export interface FrequencyRule {
+    id: string;
+    /** The window budget — the count INCLUDES the claim under evaluation (spec §13.3). */
+    max: number;
+    /** CALENDAR periods, UTC; weeks are ISO weeks (§13.2 — the chain's clock picks the window). */
+    per: FrequencyPeriod;
+    /** Informational: the engine rejects every breach regardless (§14) — copied for the record. */
+    onBreach: "review" | "reject";
+}
+/** eval-engine `schema.ts` `Unique` defaults, verbatim — see the header. */
+export declare const UNIQUE_DEFAULTS: {
+    readonly countPrior: readonly ["approved", "pending_review", "queued"];
+    readonly attachments: true;
+    readonly ai: {
+        readonly enabled: true;
+        readonly maxPairs: 3;
+    };
+    readonly thresholds: {
+        readonly rejectAtBps: 9000;
+        readonly reviewAtBps: 5000;
+    };
+};
+/** eval-engine `schema.ts` `Frequency.onBreach` default, verbatim — `reject` since engine #82
+ *  (a breach is a final reject either way; the default now says so). */
+export declare const FREQUENCY_ON_BREACH_DEFAULT: "reject";
+export declare function uniqueRuleOf(declared: ManifestUnique): UniqueRule;
+/** Null when the manifest carries no `max`/`per` — the rule cannot run (header). */
+export declare function frequencyRuleOf(declared: ManifestFrequency): FrequencyRule | null;
+//# sourceMappingURL=rule-defaults.d.ts.map

package/dist/lib/facts/stock/rule-defaults.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"rule-defaults.d.ts","sourceRoot":"","sources":["../../../../src/lib/facts/stock/rule-defaults.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,KAAK,EAAE,iBAAiB,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AAE5E,4GAA4G;AAC5G,MAAM,MAAM,QAAQ,GAAG,MAAM,GAAG;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,KAAK,GAAG,OAAO,GAAG,MAAM,CAAA;CAAE,CAAC;AAExF,MAAM,WAAW,UAAU;IACzB,EAAE,EAAE,MAAM,CAAC;IACX,4FAA4F;IAC5F,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC;IAChB,2FAA2F;IAC3F,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,+EAA+E;IAC/E,WAAW,EAAE,OAAO,CAAC;IACrB,uEAAuE;IACvE,EAAE,EAAE;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC;IAC3C,8FAA8F;IAC9F,UAAU,EAAE;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;CAC1D;AAED,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,MAAM,GAAG,OAAO,GAAG,MAAM,CAAC;AAEhE,MAAM,WAAW,aAAa;IAC5B,EAAE,EAAE,MAAM,CAAC;IACX,sFAAsF;IACtF,GAAG,EAAE,MAAM,CAAC;IACZ,+FAA+F;IAC/F,GAAG,EAAE,eAAe,CAAC;IACrB,+FAA+F;IAC/F,QAAQ,EAAE,QAAQ,GAAG,QAAQ,CAAC;CAC/B;AAED,4EAA4E;AAC5E,eAAO,MAAM,eAAe;aAC1B,UAAU,YAAG,UAAU,EAAE,gBAAgB,EAAE,QAAQ;aACnD,WAAW;aACX,EAAE;iBAAI,OAAO;iBAAQ,QAAQ,EAAE,CAAC;;aAChC,UAAU;iBAAI,WAAW,EAAE,IAAI;iBAAE,WAAW,EAAE,IAAI;;CAC1C,CAAC;AAEX;wEACwE;AACxE,eAAO,MAAM,2BAA2B,EAAG,QAAiB,CAAC;AAE7D,wBAAgB,YAAY,CAAC,QAAQ,EAAE,cAAc,GAAG,UAAU,CAejE;AAED,oFAAoF;AACpF,wBAAgB,eAAe,CAAC,QAAQ,EAAE,iBAAiB,GAAG,aAAa,GAAG,IAAI,CAQjF"}

package/dist/lib/facts/stock/rule-defaults.js

@@ -0,0 +1,55 @@
+/**
+ * The duplicate/frequency rules as the STOCK producer runs them — the engine's manifest
+ * declaration (lib/engine/types.ts `ManifestUnique` / `ManifestFrequency`, every parameter
+ * optional on the wire) resolved to a complete rule.
+ *
+ * THE defaults below mirror eval-engine `src/lib/rubric/schema.ts` `Unique` / `Frequency`
+ * (`countPrior`, `attachments`, `ai`, `thresholds`, `onBreach` — what `Unique.parse` /
+ * `Frequency.parse` yield when the author left a field out) and must be kept in LOCK-STEP with
+ * them until every engine sends the full post-default declaration (decision-engine @ 444aeae
+ * already does; an older engine sends `{ ruleId }` only). One place, on purpose: a default that
+ * drifted from the engine's would score a claim against numbers the rubric's author never set.
+ *
+ * `frequency` is different: `max` and `per` HAVE no default — a rule without them cannot run, so
+ * `frequencyRuleOf` answers null and the stock producer produces nothing; validate.ts then
+ * refuses the build (`FACT_MISSING /facts/frequency "manifest lacks max/per"` ⇒ `invalid_facts`,
+ * our/the engine's bug, never the claim's).
+ */
+/** eval-engine `schema.ts` `Unique` defaults, verbatim — see the header. */
+export const UNIQUE_DEFAULTS = {
+    countPrior: ["approved", "pending_review", "queued"],
+    attachments: true,
+    ai: { enabled: true, maxPairs: 3 },
+    thresholds: { rejectAtBps: 9000, reviewAtBps: 5000 },
+};
+/** eval-engine `schema.ts` `Frequency.onBreach` default, verbatim — `reject` since engine #82
+ *  (a breach is a final reject either way; the default now says so). */
+export const FREQUENCY_ON_BREACH_DEFAULT = "reject";
+export function uniqueRuleOf(declared) {
+    return {
+        id: declared.ruleId,
+        ...(declared.by ? { by: declared.by } : {}),
+        countPrior: declared.countPrior ?? [...UNIQUE_DEFAULTS.countPrior],
+        attachments: declared.attachments ?? UNIQUE_DEFAULTS.attachments,
+        ai: {
+            enabled: declared.ai?.enabled ?? UNIQUE_DEFAULTS.ai.enabled,
+            maxPairs: declared.ai?.maxPairs ?? UNIQUE_DEFAULTS.ai.maxPairs,
+        },
+        thresholds: {
+            rejectAtBps: declared.thresholds?.rejectAtBps ?? UNIQUE_DEFAULTS.thresholds.rejectAtBps,
+            reviewAtBps: declared.thresholds?.reviewAtBps ?? UNIQUE_DEFAULTS.thresholds.reviewAtBps,
+        },
+    };
+}
+/** Null when the manifest carries no `max`/`per` — the rule cannot run (header). */
+export function frequencyRuleOf(declared) {
+    if (declared.max === undefined || declared.per === undefined)
+        return null;
+    return {
+        id: declared.ruleId,
+        max: declared.max,
+        per: declared.per,
+        onBreach: declared.onBreach ?? FREQUENCY_ON_BREACH_DEFAULT,
+    };
+}
+//# sourceMappingURL=rule-defaults.js.map

package/dist/lib/facts/stock/rule-defaults.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"rule-defaults.js","sourceRoot":"","sources":["../../../../src/lib/facts/stock/rule-defaults.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAiCH,4EAA4E;AAC5E,MAAM,CAAC,MAAM,eAAe,GAAG;IAC7B,UAAU,EAAE,CAAC,UAAU,EAAE,gBAAgB,EAAE,QAAQ,CAAC;IACpD,WAAW,EAAE,IAAI;IACjB,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC,EAAE;IAClC,UAAU,EAAE,EAAE,WAAW,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE;CAC5C,CAAC;AAEX;wEACwE;AACxE,MAAM,CAAC,MAAM,2BAA2B,GAAG,QAAiB,CAAC;AAE7D,MAAM,UAAU,YAAY,CAAC,QAAwB;IACnD,OAAO;QACL,EAAE,EAAE,QAAQ,CAAC,MAAM;QACnB,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,QAAQ,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3C,UAAU,EAAE,QAAQ,CAAC,UAAU,IAAI,CAAC,GAAG,eAAe,CAAC,UAAU,CAAC;QAClE,WAAW,EAAE,QAAQ,CAAC,WAAW,IAAI,eAAe,CAAC,WAAW;QAChE,EAAE,EAAE;YACF,OAAO,EAAE,QAAQ,CAAC,EAAE,EAAE,OAAO,IAAI,eAAe,CAAC,EAAE,CAAC,OAAO;YAC3D,QAAQ,EAAE,QAAQ,CAAC,EAAE,EAAE,QAAQ,IAAI,eAAe,CAAC,EAAE,CAAC,QAAQ;SAC/D;QACD,UAAU,EAAE;YACV,WAAW,EAAE,QAAQ,CAAC,UAAU,EAAE,WAAW,IAAI,eAAe,CAAC,UAAU,CAAC,WAAW;YACvF,WAAW,EAAE,QAAQ,CAAC,UAAU,EAAE,WAAW,IAAI,eAAe,CAAC,UAAU,CAAC,WAAW;SACxF;KACF,CAAC;AACJ,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,eAAe,CAAC,QAA2B;IACzD,IAAI,QAAQ,CAAC,GAAG,KAAK,SAAS,IAAI,QAAQ,CAAC,GAAG,KAAK,SAAS;QAAE,OAAO,IAAI,CAAC;IAC1E,OAAO;QACL,EAAE,EAAE,QAAQ,CAAC,MAAM;QACnB,GAAG,EAAE,QAAQ,CAAC,GAAG;QACjB,GAAG,EAAE,QAAQ,CAAC,GAAG;QACjB,QAAQ,EAAE,QAAQ,CAAC,QAAQ,IAAI,2BAA2B;KAC3D,CAAC;AACJ,CAAC"}