@mailwoman/corpus 9.4.0 → 10.0.0

package/out/kr/tools/fetch/localdata.js

@@ -0,0 +1,139 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   Fetch Korea's local-government permit registry (지방행정인허가데이터): one CSV per permit category,
+ *   every business a local authority has licensed, about 195 categories. Each row carries BOTH address
+ *   forms of the same premises — `소재지전체주소` (the lot-number form, 지번) and `도로명전체주소` (the
+ *   road-name form) — plus both postcodes and a planar coordinate (`좌표정보(x/y)`, EPSG:5174). That
+ *   pairing is the cheapest dual-format signal Korean addresses offer, and the coordinate is the
+ *   board's coordinate half, which is why this is the NOISY source for Korean and not a side dish.
+ *
+ *   The public-data portal labels the category files "이용허락범위 제한 없음" and links them to
+ *   `file.localdata.go.kr`, which serves them behind a session: the category page sets the XSRF cookie,
+ *   `/file/validate/download-count` is the portal's own rate check (429 when it wants a pause), and
+ *   `/file/download/<slug>/info` streams the CSV. The files are CP949 as delivered; the adapter
+ *   decodes. The restaurant category alone is about 700 MB.
+ *
+ *   Invoke via `mailwoman corpus fetch localdata-kr --out-root <path>`, optionally with
+ *   `--categories general_restaurants,rest_cafes`.
+ */
+import { BYTES_PER_KIB } from "@mailwoman/core/fs/formatters";
+import { pathExists } from "@mailwoman/core/fs/readers";
+import { makeDirectories } from "@mailwoman/core/fs/writers";
+import { sha256File } from "@mailwoman/core/hash";
+import { sleep } from "@mailwoman/core/utils/sleep";
+import { join } from "path-ts";
+import { cookieHeader, loadCollectionFiles, streamBodyToFile, withRetries, writeManifest, } from "#tools/fetch/download/index";
+const SLUG = "localdata-kr";
+const PORTAL = "https://file.localdata.go.kr";
+const INDEX_URL = `${PORTAL}/file/general_restaurants/info`;
+const LICENSE = "이용허락범위 제한 없음 (data.go.kr label on each category file, e.g. 행정안전부_식품_일반음식점)";
+const ATTRIBUTION = "행정안전부 (Ministry of the Interior and Safety), 지방행정인허가데이터";
+const HTTP_TOO_MANY_REQUESTS = 429;
+const RATE_PAUSE_MS = 30_000;
+const USER_AGENT = "Mozilla/5.0 (X11; Linux x86_64) mailwoman-corpus-fetch";
+/**
+ * Open a session on the portal: the category page answers the XSRF cookie and the navigation that names every category.
+ */
+async function openSession() {
+    const res = await fetch(INDEX_URL, {
+        headers: { "user-agent": USER_AGENT, accept: "text/html", referer: "https://www.data.go.kr/" },
+    });
+    if (!res.ok)
+        throw new Error(`localdata-kr: the category page answered HTTP ${res.status}`);
+    const html = await res.text();
+    const cookie = cookieHeader(res);
+    const xsrf = /XSRF-TOKEN=([^;]+)/.exec(cookie)?.[1];
+    const categories = [];
+    const seen = new Set();
+    for (const [, slug, name] of html.matchAll(/href="\/file\/([a-z0-9_]+)\/info"[^>]*>\s*<span>([^<]+)<\/span>/g)) {
+        if (!slug || !name || seen.has(slug))
+            continue;
+        seen.add(slug);
+        categories.push({ slug, name: name.trim() });
+    }
+    return { cookie, xsrf, categories };
+}
+/**
+ * The portal's own rate check; a 429 carries the pause it asks for in prose, so the caller sleeps and retries.
+ */
+async function validateDownloadCount(session) {
+    const headers = { "user-agent": USER_AGENT, cookie: session.cookie, referer: INDEX_URL };
+    if (session.xsrf) {
+        headers["X-XSRF-TOKEN"] = session.xsrf;
+    }
+    const res = await fetch(`${PORTAL}/file/validate/download-count`, { headers });
+    return res.status !== HTTP_TOO_MANY_REQUESTS;
+}
+export async function fetchLocaldataKR(options, report) {
+    const destDir = join(options.outRoot, SLUG);
+    await makeDirectories(destDir);
+    const manifestPath = join(destDir, "MANIFEST.json");
+    const session = await openSession();
+    const wanted = options.categories?.length
+        ? session.categories.filter((category) => options.categories.includes(category.slug))
+        : session.categories;
+    report?.(`=== ${SLUG}: ${wanted.length} of ${session.categories.length} categories`);
+    // A previous run's entries survive, keyed by file name, so a re-run after an interruption fetches only what is missing.
+    const previous = await loadCollectionFiles(manifestPath);
+    const files = new Map();
+    let fetched = 0;
+    let skipped = 0;
+    const failedCodes = [];
+    for (const category of wanted) {
+        const filename = `${category.slug}.csv`;
+        const dest = join(destDir, filename);
+        const url = `${PORTAL}/file/download/${category.slug}/info`;
+        const before = previous.get(filename);
+        if (before && (await pathExists(dest))) {
+            files.set(filename, before);
+            skipped++;
+            continue;
+        }
+        report?.(`--- ${category.name} (${category.slug})`);
+        while (!(await validateDownloadCount(session))) {
+            report?.(`  portal asks for a pause — sleeping ${RATE_PAUSE_MS / 1000}s`);
+            await sleep(RATE_PAUSE_MS);
+        }
+        let bytes;
+        try {
+            bytes = await withRetries(async () => {
+                const res = await fetch(url, {
+                    headers: { "user-agent": USER_AGENT, cookie: session.cookie, referer: INDEX_URL, accept: "*/*" },
+                    signal: AbortSignal.timeout(1_800_000),
+                });
+                if (!res.ok || !(res.headers.get("content-type") ?? "").includes("text/csv")) {
+                    throw new Error(`HTTP ${res.status} ${res.headers.get("content-type") ?? ""} for ${category.slug}`);
+                }
+                return streamBodyToFile(res, dest);
+            }, { report, label: category.slug });
+        }
+        catch (error) {
+            report?.(`  ✗ ${error.message}`);
+            failedCodes.push(category.slug);
+            continue;
+        }
+        if (bytes < BYTES_PER_KIB) {
+            report?.(`  ✗ ${bytes} bytes — an error page, not the register`);
+            failedCodes.push(category.slug);
+            continue;
+        }
+        const sha = await sha256File(dest);
+        files.set(filename, { source_url: url, downloaded_at: new Date().toISOString(), filename, sha256: sha, bytes });
+        fetched++;
+        report?.(`  ✓ ${(bytes / 1024 / 1024).toFixed(1)} MB  sha256=${sha}`);
+        const manifest = {
+            source: SLUG,
+            source_url: INDEX_URL,
+            license: LICENSE,
+            attribution: ATTRIBUTION,
+            downloaded_at: new Date().toISOString(),
+            files: [...files.values()],
+        };
+        await writeManifest(manifestPath, manifest);
+    }
+    return { fetched, skipped, failed: failedCodes.length, failedCodes };
+}
+//# sourceMappingURL=localdata.js.map

package/out/kr/tools/fetch/localdata.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"localdata.js","sourceRoot":"","sources":["../../../../lib/kr/tools/fetch/localdata.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,+BAA+B,CAAA;AAC7D,OAAO,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAA;AACvD,OAAO,EAAE,eAAe,EAAE,MAAM,4BAA4B,CAAA;AAC5D,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAA;AACjD,OAAO,EAAE,KAAK,EAAE,MAAM,6BAA6B,CAAA;AACnD,OAAO,EAAE,IAAI,EAAE,MAAM,SAAS,CAAA;AAQ9B,OAAO,EACN,YAAY,EACZ,mBAAmB,EACnB,gBAAgB,EAChB,WAAW,EACX,aAAa,GACb,MAAM,6BAA6B,CAAA;AAEpC,MAAM,IAAI,GAAG,cAAc,CAAA;AAC3B,MAAM,MAAM,GAAG,8BAA8B,CAAA;AAC7C,MAAM,SAAS,GAAG,GAAG,MAAM,gCAAgC,CAAA;AAC3D,MAAM,OAAO,GAAG,4EAA4E,CAAA;AAC5F,MAAM,WAAW,GAAG,yDAAyD,CAAA;AAC7E,MAAM,sBAAsB,GAAG,GAAG,CAAA;AAClC,MAAM,aAAa,GAAG,MAAM,CAAA;AAoB5B,MAAM,UAAU,GAAG,wDAAwD,CAAA;AAE3E;;GAEG;AACH,KAAK,UAAU,WAAW;IACzB,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,SAAS,EAAE;QAClC,OAAO,EAAE,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,yBAAyB,EAAE;KAC9F,CAAC,CAAA;IAEF,IAAI,CAAC,GAAG,CAAC,EAAE;QAAE,MAAM,IAAI,KAAK,CAAC,iDAAiD,GAAG,CAAC,MAAM,EAAE,CAAC,CAAA;IAC3F,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAA;IAC7B,MAAM,MAAM,GAAG,YAAY,CAAC,GAAG,CAAC,CAAA;IAChC,MAAM,IAAI,GAAG,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAA;IACnD,MAAM,UAAU,GAAe,EAAE,CAAA;IACjC,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAE9B,KAAK,MAAM,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,IAAI,CAAC,QAAQ,CAAC,kEAAkE,CAAC,EAAE,CAAC;QAChH,IAAI,CAAC,IAAI,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,SAAQ;QAC9C,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;QACd,UAAU,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,EAAE,CAAC,CAAA;IAC7C,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,CAAA;AACpC,CAAC;AAED;;GAEG;AACH,KAAK,UAAU,qBAAqB,CAAC,OAAgB;IACpD,MAAM,OAAO,GAA2B,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,CAAA;IAEhH,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC;QAClB,OAAO,CAAC,cAAc,CAAC,GAAG,OAAO,CAAC,IAAI,CAAA;IACvC,CAAC;IAED,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,MAAM,+BAA+B,EAAE,EAAE,OAAO,EAAE,CAAC,CAAA;IAE9E,OAAO,GAAG,CAAC,MAAM,KAAK,sBAAsB,CAAA;AAC7C,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACrC,OAAgC,EAChC,MAA+B;IAE/B,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,IAAI,CAAC,CAAA;IAC3C,MAAM,eAAe,CAAC,OAAO,CAAC,CAAA;IAC9B,MAAM,YAAY,GAAG,IAAI,CAAC,OAAO,EAAE,eAAe,CAAC,CAAA;IAEnD,MAAM,OAAO,GAAG,MAAM,WAAW,EAAE,CAAA;IAEnC,MAAM,MAAM,GAAG,OAAO,CAAC,UAAU,EAAE,MAAM;QACxC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,OAAO,CAAC,UAAW,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACtF,CAAC,CAAC,OAAO,CAAC,UAAU,CAAA;IAErB,MAAM,EAAE,CAAC,OAAO,IAAI,KAAK,MAAM,CAAC,MAAM,OAAO,OAAO,CAAC,UAAU,CAAC,MAAM,aAAa,CAAC,CAAA;IAEpF,wHAAwH;IACxH,MAAM,QAAQ,GAAG,MAAM,mBAAmB,CAAC,YAAY,CAAC,CAAA;IACxD,MAAM,KAAK,GAAG,IAAI,GAAG,EAA0B,CAAA;IAC/C,IAAI,OAAO,GAAG,CAAC,CAAA;IACf,IAAI,OAAO,GAAG,CAAC,CAAA;IACf,MAAM,WAAW,GAAa,EAAE,CAAA;IAEhC,KAAK,MAAM,QAAQ,IAAI,MAAM,EAAE,CAAC;QAC/B,MAAM,QAAQ,GAAG,GAAG,QAAQ,CAAC,IAAI,MAAM,CAAA;QACvC,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAA;QACpC,MAAM,GAAG,GAAG,GAAG,MAAM,kBAAkB,QAAQ,CAAC,IAAI,OAAO,CAAA;QAC3D,MAAM,MAAM,GAAG,QAAQ,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAA;QAErC,IAAI,MAAM,IAAI,CAAC,MAAM,UAAU,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;YACxC,KAAK,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAA;YAE3B,OAAO,EAAE,CAAA;YAET,SAAQ;QACT,CAAC;QAED,MAAM,EAAE,CAAC,OAAO,QAAQ,CAAC,IAAI,KAAK,QAAQ,CAAC,IAAI,GAAG,CAAC,CAAA;QAEnD,OAAO,CAAC,CAAC,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC;YAChD,MAAM,EAAE,CAAC,wCAAwC,aAAa,GAAG,IAAI,GAAG,CAAC,CAAA;YACzE,MAAM,KAAK,CAAC,aAAa,CAAC,CAAA;QAC3B,CAAC;QAED,IAAI,KAAa,CAAA;QAEjB,IAAI,CAAC;YACJ,KAAK,GAAG,MAAM,WAAW,CACxB,KAAK,IAAI,EAAE;gBACV,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE;oBAC5B,OAAO,EAAE,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE;oBAChG,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,SAAS,CAAC;iBACtC,CAAC,CAAA;gBAEF,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,EAAE,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,EAAE,CAAC;oBAC9E,MAAM,IAAI,KAAK,CAAC,QAAQ,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,EAAE,QAAQ,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAA;gBACpG,CAAC;gBAED,OAAO,gBAAgB,CAAC,GAAG,EAAE,IAAI,CAAC,CAAA;YACnC,CAAC,EACD,EAAE,MAAM,EAAE,KAAK,EAAE,QAAQ,CAAC,IAAI,EAAE,CAChC,CAAA;QACF,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,EAAE,CAAC,OAAQ,KAAe,CAAC,OAAO,EAAE,CAAC,CAAA;YAC3C,WAAW,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAA;YAE/B,SAAQ;QACT,CAAC;QAED,IAAI,KAAK,GAAG,aAAa,EAAE,CAAC;YAC3B,MAAM,EAAE,CAAC,OAAO,KAAK,0CAA0C,CAAC,CAAA;YAChE,WAAW,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAA;YAE/B,SAAQ;QACT,CAAC;QAED,MAAM,GAAG,GAAG,MAAM,UAAU,CAAC,IAAI,CAAC,CAAA;QAClC,KAAK,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,aAAa,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC,CAAA;QAE/G,OAAO,EAAE,CAAA;QACT,MAAM,EAAE,CAAC,OAAO,CAAC,KAAK,GAAG,IAAI,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,GAAG,EAAE,CAAC,CAAA;QAErE,MAAM,QAAQ,GAA6B;YAC1C,MAAM,EAAE,IAAI;YACZ,UAAU,EAAE,SAAS;YACrB,OAAO,EAAE,OAAO;YAChB,WAAW,EAAE,WAAW;YACxB,aAAa,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACvC,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;SAC1B,CAAA;QAED,MAAM,aAAa,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAA;IAC5C,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,EAAE,WAAW,EAAE,CAAA;AACrE,CAAC"}

package/out/nl/recipes/postcode.d.ts

@@ -0,0 +1,26 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `nl-postcode` — the Dutch full-form postcode slice (#924). The model reads the DIGITS-FIRST NL
+ *   postcode "1012 LG" as a house number + a 2-letter street ("1012 LG Amsterdam" → house_number
+ *   1012 / street "LG" / locality Amsterdam), and that spurious street context then pulls the
+ *   locality into the US situs tier (Amsterdam → Amsterdam, NY). Letters-first postcodes (UK
+ *   "SW1A 1AA") parse natively; the `\d{4} [A-Z]{2}` shape does not, and the soft query-shape prior
+ *   (0.9 log-odds) can't overcome the strong house-number reading of a leading 4-digit token.
+ *
+ *   This is the model-first fix as DATA (the #723/#901 discipline — teach the boundary, don't
+ *   override the decoder): real NL (street, number, postcode, city) tuples in the orders Dutch
+ *   addresses actually use, with the full postcode tagged as ONE postcode span. Both the SPACED
+ *   ("1012 LG", the failing form — a 2-token span) and UNSPACED ("1012LG", 1 token) forms are
+ *   emitted so the model learns the digits-first postcode regardless of spacing; the three orders
+ *   keep polarity balanced (the v1.9.9 lesson).
+ */
+import { type CorpusRecipe } from "#recipes/scaffold";
+/**
+ * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
+export declare const nlPostcodeRecipe: CorpusRecipe;
+//# sourceMappingURL=postcode.d.ts.map

package/out/nl/recipes/postcode.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"postcode.d.ts","sourceRoot":"","sources":["../../../lib/nl/recipes/postcode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAIH,OAAO,EAA6B,KAAK,YAAY,EAAiB,MAAM,mBAAmB,CAAA;AAS/F;;;GAGG;AACH,eAAO,MAAM,gBAAgB,EAAE,YAiF9B,CAAA"}

package/out/nl/recipes/postcode.js

@@ -0,0 +1,103 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `nl-postcode` — the Dutch full-form postcode slice (#924). The model reads the DIGITS-FIRST NL
+ *   postcode "1012 LG" as a house number + a 2-letter street ("1012 LG Amsterdam" → house_number
+ *   1012 / street "LG" / locality Amsterdam), and that spurious street context then pulls the
+ *   locality into the US situs tier (Amsterdam → Amsterdam, NY). Letters-first postcodes (UK
+ *   "SW1A 1AA") parse natively; the `\d{4} [A-Z]{2}` shape does not, and the soft query-shape prior
+ *   (0.9 log-odds) can't overcome the strong house-number reading of a leading 4-digit token.
+ *
+ *   This is the model-first fix as DATA (the #723/#901 discipline — teach the boundary, don't
+ *   override the decoder): real NL (street, number, postcode, city) tuples in the orders Dutch
+ *   addresses actually use, with the full postcode tagged as ONE postcode span. Both the SPACED
+ *   ("1012 LG", the failing form — a 2-token span) and UNSPACED ("1012LG", 1 token) forms are
+ *   emitted so the model learns the digits-first postcode regardless of spacing; the three orders
+ *   keep polarity balanced (the v1.9.9 lesson).
+ */
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
+import { alignAndWrite, readTuples, sliceSourceID } from "#recipes/scaffold";
+/**
+ * "1012LG" → "1012 LG". The tuples carry the unspaced OA form; the spaced form is the failing case.
+ */
+function spacePostcode(pc) {
+    return pc.replace(/^(\d{4})([A-Z]{2})$/, "$1 $2");
+}
+/**
+ * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
+export const nlPostcodeRecipe = {
+    name: "nl-postcode",
+    description: "NL full-form postcode (#924): teach '\\d{4} [A-Z]{2}' = postcode, not house#+street — spaced + unspaced, 3 orders",
+    mode: "tuples",
+    async run(opts, write) {
+        makeMulberry32(opts.seed);
+        let read = 0;
+        let emitted = 0;
+        let skipped = 0;
+        for await (const t of readTuples(opts.input)) {
+            read++;
+            const street = String(t.street ?? "").trim();
+            const city = String(t.locality ?? "").trim();
+            const number = String(t.number ?? "").trim();
+            const rawPostcode = String(t.postcode ?? "")
+                .trim()
+                .toUpperCase()
+                .replaceAll(/\s+/g, "");
+            if (!street || !city || !number || !/^\d{4}[A-Z]{2}$/.test(rawPostcode)) {
+                skipped++;
+                continue;
+            }
+            // Spacing rotates so the model sees BOTH the failing spaced form and the unspaced form; the
+            // components.postcode value MUST match the raw form so alignment tags the right span.
+            const spaced = read % 2 === 0;
+            const postcode = spaced ? spacePostcode(rawPostcode) : rawPostcode;
+            // The three orders Dutch addresses use. `street number, postcode city` is canonical; the
+            // pc-first form is where the leading digits most strongly mis-read as a house number.
+            const order = read % 3;
+            let raw;
+            if (order === 0) {
+                raw = `${street} ${number}, ${postcode} ${city}`;
+            }
+            else if (order === 1) {
+                raw = `${postcode} ${city}, ${street} ${number}`;
+            }
+            else {
+                raw = `${city}, ${postcode}, ${street} ${number}`;
+            }
+            const components = {
+                street,
+                house_number: number,
+                postcode,
+                locality: city,
+            };
+            const source_id = sliceSourceID("synth-nl-postcode", {
+                ...components,
+                o: String(order),
+                s: spaced ? "1" : "0",
+                v: String(read),
+            });
+            const canonical = {
+                raw,
+                components,
+                country: "NL",
+                locale: "nl-NL",
+                source: "synth-nl-postcode",
+                source_id,
+                corpus_version: "0.10.0",
+                license: "Synthetic — nl-postcode; (street, number, postcode, city) from OpenAddresses NL (per-source attribution in the model card)",
+            };
+            if (alignAndWrite(write, canonical, "nl-postcode")) {
+                emitted++;
+            }
+            else {
+                skipped++;
+            }
+        }
+        return { read, emitted, skipped };
+    },
+};
+//# sourceMappingURL=postcode.js.map

package/out/nl/recipes/postcode.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"postcode.js","sourceRoot":"","sources":["../../../lib/nl/recipes/postcode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AAEpE,OAAO,EAAE,aAAa,EAAE,UAAU,EAAqB,aAAa,EAAE,MAAM,mBAAmB,CAAA;AAE/F;;GAEG;AACH,SAAS,aAAa,CAAC,EAAU;IAChC,OAAO,EAAE,CAAC,OAAO,CAAC,qBAAqB,EAAE,OAAO,CAAC,CAAA;AAClD,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAiB;IAC7C,IAAI,EAAE,aAAa;IACnB,WAAW,EACV,mHAAmH;IACpH,IAAI,EAAE,QAAQ;IACd,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACzB,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QAEf,IAAI,KAAK,EAAE,MAAM,CAAC,IAAI,UAAU,CAAC,IAAI,CAAC,KAAM,CAAC,EAAE,CAAC;YAC/C,IAAI,EAAE,CAAA;YACN,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;YAC5C,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;YAC5C,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;YAE5C,MAAM,WAAW,GAAG,MAAM,CAAC,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC;iBAC1C,IAAI,EAAE;iBACN,WAAW,EAAE;iBACb,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC,CAAA;YAExB,IAAI,CAAC,MAAM,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,iBAAiB,CAAC,IAAI,CAAC,WAAW,CAAC,EAAE,CAAC;gBACzE,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,4FAA4F;YAC5F,sFAAsF;YACtF,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,CAAA;YAC7B,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,aAAa,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,WAAW,CAAA;YAElE,yFAAyF;YACzF,sFAAsF;YACtF,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAA;YACtB,IAAI,GAAW,CAAA;YAEf,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACjB,GAAG,GAAG,GAAG,MAAM,IAAI,MAAM,KAAK,QAAQ,IAAI,IAAI,EAAE,CAAA;YACjD,CAAC;iBAAM,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACxB,GAAG,GAAG,GAAG,QAAQ,IAAI,IAAI,KAAK,MAAM,IAAI,MAAM,EAAE,CAAA;YACjD,CAAC;iBAAM,CAAC;gBACP,GAAG,GAAG,GAAG,IAAI,KAAK,QAAQ,KAAK,MAAM,IAAI,MAAM,EAAE,CAAA;YAClD,CAAC;YAED,MAAM,UAAU,GAA2B;gBAC1C,MAAM;gBACN,YAAY,EAAE,MAAM;gBACpB,QAAQ;gBACR,QAAQ,EAAE,IAAI;aACd,CAAA;YAED,MAAM,SAAS,GAAG,aAAa,CAAC,mBAAmB,EAAE;gBACpD,GAAG,UAAU;gBACb,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC;gBAChB,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG;gBACrB,CAAC,EAAE,MAAM,CAAC,IAAI,CAAC;aACf,CAAC,CAAA;YAEF,MAAM,SAAS,GAAG;gBACjB,GAAG;gBACH,UAAU;gBACV,OAAO,EAAE,IAAI;gBACb,MAAM,EAAE,OAAO;gBACf,MAAM,EAAE,mBAAmB;gBAC3B,SAAS;gBACT,cAAc,EAAE,QAAQ;gBACxB,OAAO,EACN,4HAA4H;aAC7H,CAAA;YAED,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,EAAE,aAAa,CAAC,EAAE,CAAC;gBACpD,OAAO,EAAE,CAAA;YACV,CAAC;iBAAM,CAAC;gBACP,OAAO,EAAE,CAAA;YACV,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}

package/out/no/recipes/fragment.d.ts

@@ -0,0 +1,43 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `no-fragment` — the Norwegian house-number-licence change (Track B, 2026-07-16). The mirror of
+ *   `fr-fragment`, which earned +50pp on the same defect shape in French.
+ *
+ *   WHY THIS EXISTS AND `no-street-led` DOES NOT SUFFICE. Board 3 (the NO digit board) measured that
+ *   `synth-no-street-led`'s three forms — all carrying postcode+city — are ALREADY at 0.940-0.968 on
+ *   a model with zero Norwegian rows. The headroom is in the forms that slice never emits:
+ *
+ *     bare-street-hn   "Hallingrudveien 32"     0.693   — no postcode competing, still fails 31%
+ *     slash-hn         "Øvrabø 124/1"           0.650   — cadastral gnr/bnr, ONE component
+ *
+ *   THE MECHANISM the board exposed: `Hallingrudveien 32` -> locality + postcode, while
+ *   `Hallingrudveien 32, 3370 Vikersund` parses perfectly. The street loses its street reading and
+ *   the digit loses its anchor TOGETHER. That is Track A's bare-street LICENCE in Norwegian — the
+ *   model will not read a street without its postcode/locality partner — not a digit-ownership prior.
+ *   fr-fragment fixed exactly this in French by teaching the street WITHOUT its partners.
+ *
+ *   THE COUNTER-DISTRIBUTION IS THE POINT (fr-fragment's lesson, and board 2's bare-locality guard).
+ *   Teaching bare `{street} {number}` alone lets the model satisfy every row by flipping its default
+ *   from "bare toponym -> locality" to "bare toponym -> street", trading one broken prior for
+ *   another. Two counter-classes hold the line:
+ *     - bare LOCALITIES (no street) so "bare -> street" is not free.
+ *     - bare POSTCODES so the model does not learn to stop emitting postcode to win the digit — the
+ *       board 3 bare-pc negative class (1.000) must HOLD.
+ *
+ *   SLASH HAZARD, pinned deliberately: NO `124/1` is ONE house_number (cadastral gnr/bnr). AU
+ *   `12/345` is TWO (unit + house_number). This slice teaches the Norwegian reading; a future AU
+ *   intra-word-split slice (B5) must not generalize over it. The two are locale-restricted by design.
+ *
+ *   SPLIT: `--exclude-surfaces` REQUIRED (throws otherwise) — the digit board's reserved surface
+ *   list. Diacritic-KEEPING normalizer, matching the board (see the norm docstring).
+ */
+import { type CorpusRecipe } from "#recipes/scaffold";
+/**
+ * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
+export declare const noFragmentRecipe: CorpusRecipe;
+//# sourceMappingURL=fragment.d.ts.map

package/out/no/recipes/fragment.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"fragment.d.ts","sourceRoot":"","sources":["../../../lib/no/recipes/fragment.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmCG;AAKH,OAAO,EAA4C,KAAK,YAAY,EAAiB,MAAM,mBAAmB,CAAA;AAW9G;;;GAGG;AACH,eAAO,MAAM,gBAAgB,EAAE,YA6J9B,CAAA"}

package/out/no/recipes/fragment.js

@@ -0,0 +1,183 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `no-fragment` — the Norwegian house-number-licence change (Track B, 2026-07-16). The mirror of
+ *   `fr-fragment`, which earned +50pp on the same defect shape in French.
+ *
+ *   WHY THIS EXISTS AND `no-street-led` DOES NOT SUFFICE. Board 3 (the NO digit board) measured that
+ *   `synth-no-street-led`'s three forms — all carrying postcode+city — are ALREADY at 0.940-0.968 on
+ *   a model with zero Norwegian rows. The headroom is in the forms that slice never emits:
+ *
+ *     bare-street-hn   "Hallingrudveien 32"     0.693   — no postcode competing, still fails 31%
+ *     slash-hn         "Øvrabø 124/1"           0.650   — cadastral gnr/bnr, ONE component
+ *
+ *   THE MECHANISM the board exposed: `Hallingrudveien 32` -> locality + postcode, while
+ *   `Hallingrudveien 32, 3370 Vikersund` parses perfectly. The street loses its street reading and
+ *   the digit loses its anchor TOGETHER. That is Track A's bare-street LICENCE in Norwegian — the
+ *   model will not read a street without its postcode/locality partner — not a digit-ownership prior.
+ *   fr-fragment fixed exactly this in French by teaching the street WITHOUT its partners.
+ *
+ *   THE COUNTER-DISTRIBUTION IS THE POINT (fr-fragment's lesson, and board 2's bare-locality guard).
+ *   Teaching bare `{street} {number}` alone lets the model satisfy every row by flipping its default
+ *   from "bare toponym -> locality" to "bare toponym -> street", trading one broken prior for
+ *   another. Two counter-classes hold the line:
+ *     - bare LOCALITIES (no street) so "bare -> street" is not free.
+ *     - bare POSTCODES so the model does not learn to stop emitting postcode to win the digit — the
+ *       board 3 bare-pc negative class (1.000) must HOLD.
+ *
+ *   SLASH HAZARD, pinned deliberately: NO `124/1` is ONE house_number (cadastral gnr/bnr). AU
+ *   `12/345` is TWO (unit + house_number). This slice teaches the Norwegian reading; a future AU
+ *   intra-word-split slice (B5) must not generalize over it. The two are locale-restricted by design.
+ *
+ *   SPLIT: `--exclude-surfaces` REQUIRED (throws otherwise) — the digit board's reserved surface
+ *   list. Diacritic-KEEPING normalizer, matching the board (see the norm docstring).
+ */
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
+import { TextSpliterator } from "spliterator";
+import { alignAndWrite, foldNOSurface, readTuples, sliceSourceID } from "#recipes/scaffold";
+/**
+ * Title-case a Kartverket ALL-CAPS locality (HELLVIK -> Hellvik); #690, all-caps is OOD.
+ */
+const titleNO = (value) => value
+    .split(/\s+/)
+    .map((w) => (w ? w[0].toUpperCase() + w.slice(1).toLowerCase() : w))
+    .join(" ");
+/**
+ * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
+export const noFragmentRecipe = {
+    name: "no-fragment",
+    description: "NO street fragments — the house-number-licence change (Track B): '«st» «n»' / bare «st» with NO postcode partner, + bare-locality & bare-postcode counters",
+    mode: "tuples",
+    options: [
+        {
+            flag: "--exclude-surfaces <path>",
+            description: "REQUIRED. The NO digit board's reserved surface list (mailwoman/eval-harness/fixtures/no-digits.surfaces.txt).",
+        },
+        {
+            flag: "--bare-street-prob <n>",
+            description: "Share of street rows emitted as a BARE street, no number (default 0.30 — the pure licence signal)",
+        },
+        {
+            flag: "--counter-prob <n>",
+            description: "Share of ALL rows that are counter-distribution (bare locality OR bare postcode) (default 0.30)",
+        },
+        {
+            flag: "--long-number-boost <n>",
+            description: "knob 3: emit N copies of each street+number row whose number has >= --long-number-min-digits digits — oversample the failing long-number class the model calls a postcode (default 1 = no boost)",
+        },
+        {
+            flag: "--long-number-min-digits <n>",
+            description: "knob 3: minimum digit count for a number to be 'long' and boosted (default 3)",
+        },
+    ],
+    async run(opts, write) {
+        const random = makeMulberry32(opts.seed);
+        const excludePath = opts.excludeSurfaces;
+        if (!excludePath) {
+            throw new Error("no-fragment: --exclude-surfaces is REQUIRED. Pass the NO digit board's reserved list " +
+                "(mailwoman/eval-harness/fixtures/no-digits.surfaces.txt) or this slice trains on its own eval set. " +
+                "Source-disjoint by street SURFACE is the split discipline; there is no safe default.");
+        }
+        const excluded = new Set();
+        for await (const line of TextSpliterator.fromAsync(excludePath)) {
+            const trimmed = line.trim();
+            if (trimmed && !trimmed.startsWith("#")) {
+                excluded.add(trimmed);
+            }
+        }
+        if (!excluded.size)
+            throw new Error(`no-fragment: --exclude-surfaces "${excludePath}" listed no surfaces`);
+        const bareStreetProb = opts.bareProb ?? 0.3;
+        const counterProb = opts.counterProb ?? 0.3;
+        const longNumberBoost = Math.max(1, Math.floor(opts.longNumberBoost ?? 1));
+        const longNumberMinDigits = opts.longNumberMinDigits ?? 3;
+        // Harvested from the tuples — every NO row carries its locality and postcode, so the two
+        // counter-classes need no second source.
+        const localities = new Set();
+        const postcodes = new Set();
+        let read = 0;
+        let emitted = 0;
+        let skipped = 0;
+        let contaminated = 0;
+        let emitSeq = 0;
+        const emit = (raw, components, klass) => {
+            // emitSeq keeps every emit distinct — knob 3 emits N copies of one long-number row, and
+            // (components, read) alone would collide their source_id and let downstream dedup drop the boost.
+            const source_id = sliceSourceID("synth-no-fragment", { ...components, k: klass, v: `${read}:${emitSeq++}` });
+            const canonical = {
+                raw,
+                components,
+                country: "NO",
+                locale: "nb-NO",
+                source: "synth-no-fragment",
+                source_id,
+                corpus_version: "0.11.0",
+                license: "Synthetic — no-fragment; (street, number, postcode, city) from OpenAddresses NO / Kartverket",
+            };
+            if (alignAndWrite(write, canonical, "no-fragment")) {
+                emitted++;
+            }
+            else {
+                skipped++;
+            }
+        };
+        for await (const tuple of readTuples(opts.input)) {
+            read++;
+            const street = String(tuple.street ?? "").trim();
+            const locality = titleNO(String(tuple.locality ?? "").trim());
+            const number = String(tuple.number ?? "").trim();
+            const postcode = String(tuple.postcode ?? "").trim();
+            if (locality) {
+                localities.add(locality);
+            }
+            if (postcode) {
+                postcodes.add(postcode);
+            }
+            if (!street) {
+                skipped++;
+                continue;
+            }
+            // THE SPLIT. A surface on the digit board never enters training.
+            if (excluded.has(foldNOSurface(street))) {
+                contaminated++;
+                continue;
+            }
+            // COUNTER-DISTRIBUTION — drawn from the harvested pools, not this row's street. Half bare
+            // localities (so "bare -> street" is not free), half bare postcodes (so the model does not
+            // stop emitting postcode to win the digit — board 3's bare-pc must hold).
+            if (random() < counterProb) {
+                if (random() < 0.5 && localities.size) {
+                    const loc = [...localities][Math.floor(random() * localities.size)];
+                    emit(loc, { locality: loc }, "counter-bare-locality");
+                }
+                else if (postcodes.size) {
+                    const pc = [...postcodes][Math.floor(random() * postcodes.size)];
+                    emit(pc, { postcode: pc }, "counter-bare-postcode");
+                }
+                continue;
+            }
+            // THE SIGNAL. A street with NO postcode/locality partner. Either bare, or street+number —
+            // both are the forms board 3 measured as the headroom (bare-street-hn 0.693, slash-hn 0.650).
+            if (!number || random() < bareStreetProb) {
+                emit(street, { street }, "bare-street");
+            }
+            else {
+                const klass = number.includes("/") ? "slash-hn" : "street-hn";
+                // knob 3: the failing class is street + LONG number (Leppdalsvegen 1285 -> postcode). The
+                // digit count, not the slash, is what tips the length prior toward postcode. Oversample
+                // those rows to fight the prior with volume and teach the street/number boundary directly.
+                const digits = (number.match(/\d/g) ?? []).length;
+                const copies = digits >= longNumberMinDigits ? longNumberBoost : 1;
+                for (let c = 0; c < copies; c++) {
+                    emit(`${street} ${number}`, { street, house_number: number }, klass);
+                }
+            }
+        }
+        return { read, emitted, skipped, contaminated };
+    },
+};
+//# sourceMappingURL=fragment.js.map

package/out/no/recipes/fragment.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fragment.js","sourceRoot":"","sources":["../../../lib/no/recipes/fragment.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmCG;AAEH,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AACpE,OAAO,EAAE,eAAe,EAAE,MAAM,aAAa,CAAA;AAE7C,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,UAAU,EAAqB,aAAa,EAAE,MAAM,mBAAmB,CAAA;AAE9G;;GAEG;AACH,MAAM,OAAO,GAAG,CAAC,KAAa,EAAU,EAAE,CACzC,KAAK;KACH,KAAK,CAAC,KAAK,CAAC;KACZ,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAE,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;KACpE,IAAI,CAAC,GAAG,CAAC,CAAA;AAEZ;;;GAGG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAiB;IAC7C,IAAI,EAAE,aAAa;IACnB,WAAW,EACV,4JAA4J;IAC7J,IAAI,EAAE,QAAQ;IACd,OAAO,EAAE;QACR;YACC,IAAI,EAAE,2BAA2B;YACjC,WAAW,EACV,gHAAgH;SACjH;QACD;YACC,IAAI,EAAE,wBAAwB;YAC9B,WAAW,EAAE,mGAAmG;SAChH;QACD;YACC,IAAI,EAAE,oBAAoB;YAC1B,WAAW,EAAE,iGAAiG;SAC9G;QACD;YACC,IAAI,EAAE,yBAAyB;YAC/B,WAAW,EACV,kMAAkM;SACnM;QACD;YACC,IAAI,EAAE,8BAA8B;YACpC,WAAW,EAAE,+EAA+E;SAC5F;KACD;IACD,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,MAAM,MAAM,GAAG,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACxC,MAAM,WAAW,GAAG,IAAI,CAAC,eAAe,CAAA;QAExC,IAAI,CAAC,WAAW,EAAE,CAAC;YAClB,MAAM,IAAI,KAAK,CACd,uFAAuF;gBACtF,qGAAqG;gBACrG,sFAAsF,CACvF,CAAA;QACF,CAAC;QAED,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAU,CAAA;QAElC,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,eAAe,CAAC,SAAS,CAAC,WAAW,CAAC,EAAE,CAAC;YACjE,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAA;YAE3B,IAAI,OAAO,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;gBACzC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;YACtB,CAAC;QACF,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,WAAW,sBAAsB,CAAC,CAAA;QAE1G,MAAM,cAAc,GAAG,IAAI,CAAC,QAAQ,IAAI,GAAG,CAAA;QAC3C,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,GAAG,CAAA;QAC3C,MAAM,eAAe,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,eAAe,IAAI,CAAC,CAAC,CAAC,CAAA;QAC1E,MAAM,mBAAmB,GAAG,IAAI,CAAC,mBAAmB,IAAI,CAAC,CAAA;QAEzD,yFAAyF;QACzF,yCAAyC;QACzC,MAAM,UAAU,GAAG,IAAI,GAAG,EAAU,CAAA;QACpC,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAA;QAEnC,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,YAAY,GAAG,CAAC,CAAA;QACpB,IAAI,OAAO,GAAG,CAAC,CAAA;QAEf,MAAM,IAAI,GAAG,CAAC,GAAW,EAAE,UAAkC,EAAE,KAAa,EAAQ,EAAE;YACrF,wFAAwF;YACxF,kGAAkG;YAClG,MAAM,SAAS,GAAG,aAAa,CAAC,mBAAmB,EAAE,EAAE,GAAG,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,IAAI,IAAI,OAAO,EAAE,EAAE,EAAE,CAAC,CAAA;YAE5G,MAAM,SAAS,GAAG;gBACjB,GAAG;gBACH,UAAU;gBACV,OAAO,EAAE,IAAI;gBACb,MAAM,EAAE,OAAO;gBACf,MAAM,EAAE,mBAAmB;gBAC3B,SAAS;gBACT,cAAc,EAAE,QAAQ;gBACxB,OAAO,EAAE,8FAA8F;aACvG,CAAA;YAED,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,EAAE,aAAa,CAAC,EAAE,CAAC;gBACpD,OAAO,EAAE,CAAA;YACV,CAAC;iBAAM,CAAC;gBACP,OAAO,EAAE,CAAA;YACV,CAAC;QACF,CAAC,CAAA;QAED,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,UAAU,CAAC,IAAI,CAAC,KAAM,CAAC,EAAE,CAAC;YACnD,IAAI,EAAE,CAAA;YACN,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;YAChD,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAA;YAC7D,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;YAChD,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;YAEpD,IAAI,QAAQ,EAAE,CAAC;gBACd,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAA;YACzB,CAAC;YAED,IAAI,QAAQ,EAAE,CAAC;gBACd,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAA;YACxB,CAAC;YAED,IAAI,CAAC,MAAM,EAAE,CAAC;gBACb,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,iEAAiE;YACjE,IAAI,QAAQ,CAAC,GAAG,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;gBACzC,YAAY,EAAE,CAAA;gBAEd,SAAQ;YACT,CAAC;YAED,0FAA0F;YAC1F,2FAA2F;YAC3F,0EAA0E;YAC1E,IAAI,MAAM,EAAE,GAAG,WAAW,EAAE,CAAC;gBAC5B,IAAI,MAAM,EAAE,GAAG,GAAG,IAAI,UAAU,CAAC,IAAI,EAAE,CAAC;oBACvC,MAAM,GAAG,GAAG,CAAC,GAAG,UAAU,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,UAAU,CAAC,IAAI,CAAC,CAAE,CAAA;oBAEpE,IAAI,CAAC,GAAG,EAAE,EAAE,QAAQ,EAAE,GAAG,EAAE,EAAE,uBAAuB,CAAC,CAAA;gBACtD,CAAC;qBAAM,IAAI,SAAS,CAAC,IAAI,EAAE,CAAC;oBAC3B,MAAM,EAAE,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,SAAS,CAAC,IAAI,CAAC,CAAE,CAAA;oBAEjE,IAAI,CAAC,EAAE,EAAE,EAAE,QAAQ,EAAE,EAAE,EAAE,EAAE,uBAAuB,CAAC,CAAA;gBACpD,CAAC;gBAED,SAAQ;YACT,CAAC;YAED,0FAA0F;YAC1F,8FAA8F;YAC9F,IAAI,CAAC,MAAM,IAAI,MAAM,EAAE,GAAG,cAAc,EAAE,CAAC;gBAC1C,IAAI,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,EAAE,aAAa,CAAC,CAAA;YACxC,CAAC;iBAAM,CAAC;gBACP,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,WAAW,CAAA;gBAC7D,0FAA0F;gBAC1F,wFAAwF;gBACxF,2FAA2F;gBAC3F,MAAM,MAAM,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAA;gBACjD,MAAM,MAAM,GAAG,MAAM,IAAI,mBAAmB,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,CAAA;gBAElE,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;oBACjC,IAAI,CAAC,GAAG,MAAM,IAAI,MAAM,EAAE,EAAE,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,EAAE,EAAE,KAAK,CAAC,CAAA;gBACrE,CAAC;YACF,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,YAAY,EAAE,CAAA;IAChD,CAAC;CACD,CAAA"}

package/out/no/recipes/street/index.d.ts

@@ -0,0 +1,18 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `no-street` slice recipe — synthetic NO-street counter-example rows: tuples →
+ *   {@link synthesizeNoStreetRow} → aligned LabeledRow. The corpus-side counterweight to the
+ *   synth-street slice that drove v0.6.1's `dependent_locality` regression — venue+admin and
+ *   admin-only rows with explicit absence of any street-side component. Ported from
+ *   scripts/build-no-street-slice.mjs.
+ */
+import { type CorpusRecipe } from "#recipes/scaffold";
+/**
+ * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
+export declare const noStreetRecipe: CorpusRecipe;
+//# sourceMappingURL=index.d.ts.map

package/out/no/recipes/street/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../lib/no/recipes/street/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAIH,OAAO,EAKN,KAAK,YAAY,EACjB,MAAM,mBAAmB,CAAA;AAG1B;;;GAGG;AACH,eAAO,MAAM,cAAc,EAAE,YA+D5B,CAAA"}

package/out/no/recipes/street/index.js

@@ -0,0 +1,72 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `no-street` slice recipe — synthetic NO-street counter-example rows: tuples →
+ *   {@link synthesizeNoStreetRow} → aligned LabeledRow. The corpus-side counterweight to the
+ *   synth-street slice that drove v0.6.1's `dependent_locality` regression — venue+admin and
+ *   admin-only rows with explicit absence of any street-side component. Ported from
+ *   scripts/build-no-street-slice.mjs.
+ */
+import { makeLcg } from "@mailwoman/core/utils";
+import { alignAndWrite, readTuples, sliceSourceID, SYNTHETIC_TUPLE_LICENSE as LICENSE, } from "#recipes/scaffold";
+import { synthesizeNoStreetRow } from "#synthesizers/no-street";
+/**
+ * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
+export const noStreetRecipe = {
+    name: "no-street",
+    description: "No-street counter-example rows: tuples → synthesizeNoStreetRow → aligned LabeledRow",
+    mode: "tuples",
+    async run(opts, write) {
+        if (!opts.input)
+            throw new Error("no-street recipe requires --input <tuples.jsonl>");
+        // Legacy build-no-street-slice.mjs seeded the LCG via makeRandom(opts.seed) (s = seed).
+        const random = makeLcg(opts.seed);
+        const source = opts.sourceName ?? "synth-no-street";
+        let read = 0;
+        let emitted = 0;
+        let skipped = 0;
+        for await (const tuple of readTuples(opts.input)) {
+            read++;
+            if (!tuple.locality || !tuple.region || !tuple.postcode || !tuple.country) {
+                skipped++;
+                continue;
+            }
+            for (let v = 0; v < opts.variants; v++) {
+                const synth = synthesizeNoStreetRow(tuple, { random });
+                if (!synth) {
+                    skipped++;
+                    continue;
+                }
+                const ok = alignAndWrite(write, {
+                    raw: synth.raw,
+                    components: synth.components,
+                    country: tuple.country,
+                    locale: synth.locale,
+                    source,
+                    source_id: sliceSourceID(source, {
+                        locality: tuple.locality,
+                        region: tuple.region,
+                        postcode: tuple.postcode,
+                        country: tuple.country,
+                        template: synth.template,
+                        v: String(v),
+                    }),
+                    corpus_version: "0.4.0",
+                    license: LICENSE,
+                }, synth.template);
+                if (ok) {
+                    emitted++;
+                }
+                else {
+                    skipped++;
+                }
+            }
+        }
+        return { read, emitted, skipped };
+    },
+};
+//# sourceMappingURL=index.js.map

package/out/no/recipes/street/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../lib/no/recipes/street/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,uBAAuB,CAAA;AAE/C,OAAO,EACN,aAAa,EACb,UAAU,EACV,aAAa,EACb,uBAAuB,IAAI,OAAO,GAElC,MAAM,mBAAmB,CAAA;AAC1B,OAAO,EAAE,qBAAqB,EAA0B,MAAM,yBAAyB,CAAA;AAEvF;;;GAGG;AACH,MAAM,CAAC,MAAM,cAAc,GAAiB;IAC3C,IAAI,EAAE,WAAW;IACjB,WAAW,EAAE,qFAAqF;IAClG,IAAI,EAAE,QAAQ;IACd,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,IAAI,CAAC,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAA;QACpF,wFAAwF;QACxF,MAAM,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACjC,MAAM,MAAM,GAAG,IAAI,CAAC,UAAU,IAAI,iBAAiB,CAAA;QACnD,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QAEf,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAClD,IAAI,EAAE,CAAA;YAEN,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBAC3E,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;gBACxC,MAAM,KAAK,GAAG,qBAAqB,CAAC,KAA0B,EAAE,EAAE,MAAM,EAAE,CAAC,CAAA;gBAE3E,IAAI,CAAC,KAAK,EAAE,CAAC;oBACZ,OAAO,EAAE,CAAA;oBAET,SAAQ;gBACT,CAAC;gBAED,MAAM,EAAE,GAAG,aAAa,CACvB,KAAK,EACL;oBACC,GAAG,EAAE,KAAK,CAAC,GAAG;oBACd,UAAU,EAAE,KAAK,CAAC,UAAU;oBAC5B,OAAO,EAAE,KAAK,CAAC,OAAO;oBACtB,MAAM,EAAE,KAAK,CAAC,MAAM;oBACpB,MAAM;oBACN,SAAS,EAAE,aAAa,CAAC,MAAM,EAAE;wBAChC,QAAQ,EAAE,KAAK,CAAC,QAAQ;wBACxB,MAAM,EAAE,KAAK,CAAC,MAAM;wBACpB,QAAQ,EAAE,KAAK,CAAC,QAAQ;wBACxB,OAAO,EAAE,KAAK,CAAC,OAAO;wBACtB,QAAQ,EAAE,KAAK,CAAC,QAAQ;wBACxB,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;qBACZ,CAAC;oBACF,cAAc,EAAE,OAAO;oBACvB,OAAO,EAAE,OAAO;iBAChB,EACD,KAAK,CAAC,QAAQ,CACd,CAAA;gBAED,IAAI,EAAE,EAAE,CAAC;oBACR,OAAO,EAAE,CAAA;gBACV,CAAC;qBAAM,CAAC;oBACP,OAAO,EAAE,CAAA;gBACV,CAAC;YACF,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}