@mailwoman/corpus 9.3.0 → 10.0.0

package/out/cz/recipes/pcfirst-preposition.js

@@ -0,0 +1,101 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `cz-pcfirst-preposition` — the Czech pc-first prepositional-locality slice, fourth orthography
+ *   of the #901 leading-name-boundary family. The #897 close-out read all 8 residual CZ rows as one
+ *   class: a LEADING postcode mis-assigned as house_number while the multi-word "nad/pod/u X"
+ *   locality shatters ("51244 Rokytnice nad Jizerou, Dolní Rokytnice 111" → street
+ *   'RokytnicenadJizerou' + house_number '51244'). That leading-5-digit confusion is the
+ *   anchor-pollution class whose decode-time OVERRIDE was correctly killed in #723 — this slice is
+ *   the model-first fix as DATA: real prepositional localities in the order that breaks, so the
+ *   model learns that a leading postcode before a multi-word name is a postcode. pc-first leads the
+ *   cycle (the lesson); canonical and city-first keep the polarity balanced (the v1.9.9 lesson).
+ */
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
+import { alignAndWrite, readTuples, sliceSourceID } from "#recipes/scaffold";
+/**
+ * The order-cycle slot for the STREET-LESS form (`«city» «pc», Česko`) — the exact surface of the
+ * `cz-full-praha-100-00` board row, whose absence from the street-bearing orders was the v4.5.0 no-promote's measured
+ * gap.
+ */
+const STREETLESS_ORDER = 3;
+/**
+ * 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 czPcFirstPrepositionRecipe = {
+    name: "cz-pcfirst-preposition",
+    description: "CZ pc-first + prepositional locality (#901 family): '«pc» «city nad X», «st» «n»' — the #723 class as data",
+    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 postcode = String(t.postcode ?? "").trim();
+            if (!street || !city || !number || !postcode) {
+                skipped++;
+                continue;
+            }
+            const order = read % 4;
+            // The OFFICIAL Czech rendering spaces the PSČ as `NNN NN` ('512 44'); OpenAddresses stores it
+            // unspaced ('51244'), and a model trained only on the source form reads the spaced surface as
+            // house_number + garbage (the 'Praha 100 00' mangle). Alternate the two renderings so both
+            // orthographies are attested — the LABEL is the postcode either way.
+            const spaced = read % 2 === 0 && /^\d{5}$/.test(postcode);
+            const postcodeSurface = spaced ? `${postcode.slice(0, 3)} ${postcode.slice(3)}` : postcode;
+            let raw;
+            const components = order === STREETLESS_ORDER
+                ? { locality: city, postcode: postcodeSurface, country: "Česko" }
+                : { street, house_number: number, postcode: postcodeSurface, locality: city };
+            if (order === 0) {
+                raw = `${postcodeSurface} ${city}, ${street} ${number}`;
+            }
+            else if (order === 1) {
+                raw = `${street} ${number}, ${postcodeSurface} ${city}`;
+            }
+            else if (order === 2) {
+                raw = `${city}, ${postcodeSurface}, ${street} ${number}`;
+            }
+            else {
+                // order === STREETLESS_ORDER: the street-less form — `«city» «pc», Česko` — the exact surface of the
+                // cz-full-praha-100-00 board row. The v4.5.0 no-promote receipt measured the gap: every
+                // prior order was street-bearing, so the model never saw a spaced PSČ beside a bare
+                // locality and mangled 'Praha 100 00, Czechia' into house_number spans. Street/number
+                // stay OUT of the components for this form (they are not in the surface).
+                raw = `${city} ${postcodeSurface}, Česko`;
+            }
+            const source_id = sliceSourceID("synth-cz-pcfirst-preposition", {
+                ...components,
+                o: String(order),
+                s: spaced ? "1" : "0",
+                v: String(read),
+            });
+            const canonical = {
+                raw,
+                components,
+                country: "CZ",
+                locale: "cs-CZ",
+                source: "synth-cz-pcfirst-preposition",
+                source_id,
+                corpus_version: "0.10.0",
+                license: "Synthetic — cz-pcfirst-preposition; (street, number, postcode, city) from OpenAddresses CZ (per-source attribution in the model card)",
+            };
+            if (alignAndWrite(write, canonical, "cz-pcfirst-preposition")) {
+                emitted++;
+            }
+            else {
+                skipped++;
+            }
+        }
+        return { read, emitted, skipped };
+    },
+};
+//# sourceMappingURL=pcfirst-preposition.js.map

package/out/cz/recipes/pcfirst-preposition.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pcfirst-preposition.js","sourceRoot":"","sources":["../../../lib/cz/recipes/pcfirst-preposition.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AAEpE,OAAO,EAAE,aAAa,EAAE,UAAU,EAAqB,aAAa,EAAE,MAAM,mBAAmB,CAAA;AAC/F;;;;GAIG;AACH,MAAM,gBAAgB,GAAG,CAAC,CAAA;AAE1B;;;GAGG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAAiB;IACvD,IAAI,EAAE,wBAAwB;IAC9B,WAAW,EACV,4GAA4G;IAC7G,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;YAC5C,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;YAEhD,IAAI,CAAC,MAAM,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,QAAQ,EAAE,CAAC;gBAC9C,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAA;YACtB,8FAA8F;YAC9F,8FAA8F;YAC9F,2FAA2F;YAC3F,qEAAqE;YACrE,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,IAAI,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;YACzD,MAAM,eAAe,GAAG,MAAM,CAAC,CAAC,CAAC,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAA;YAC1F,IAAI,GAAW,CAAA;YAEf,MAAM,UAAU,GACf,KAAK,KAAK,gBAAgB;gBACzB,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,QAAQ,EAAE,eAAe,EAAE,OAAO,EAAE,OAAO,EAAE;gBACjE,CAAC,CAAC,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,EAAE,QAAQ,EAAE,eAAe,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAA;YAE/E,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACjB,GAAG,GAAG,GAAG,eAAe,IAAI,IAAI,KAAK,MAAM,IAAI,MAAM,EAAE,CAAA;YACxD,CAAC;iBAAM,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACxB,GAAG,GAAG,GAAG,MAAM,IAAI,MAAM,KAAK,eAAe,IAAI,IAAI,EAAE,CAAA;YACxD,CAAC;iBAAM,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACxB,GAAG,GAAG,GAAG,IAAI,KAAK,eAAe,KAAK,MAAM,IAAI,MAAM,EAAE,CAAA;YACzD,CAAC;iBAAM,CAAC;gBACP,qGAAqG;gBACrG,wFAAwF;gBACxF,oFAAoF;gBACpF,sFAAsF;gBACtF,0EAA0E;gBAC1E,GAAG,GAAG,GAAG,IAAI,IAAI,eAAe,SAAS,CAAA;YAC1C,CAAC;YAED,MAAM,SAAS,GAAG,aAAa,CAAC,8BAA8B,EAAE;gBAC/D,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,8BAA8B;gBACtC,SAAS;gBACT,cAAc,EAAE,QAAQ;gBACxB,OAAO,EACN,uIAAuI;aACxI,CAAA;YAED,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,EAAE,wBAAwB,CAAC,EAAE,CAAC;gBAC/D,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/de/recipes/locale.d.ts

@@ -0,0 +1,44 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `german` slice recipe — German coverage rows from REAL OpenAddresses tuples (Berlin + Saxony,
+ *   cached zips). Each sampled tuple is rendered via {@link synthesizeGermanRow} in BOTH orders —
+ *   `--intl-fraction` (default 0.4) in international order (house-first / postcode-after-city), the
+ *   rest in idiomatic German order — then aligned to BIO. Generate-mode: it builds a tuple pool
+ *   from the cached zips, then draws `--count` rows from it with the passed `random` (so the emit
+ *   stream matches the legacy reservoir-sample loop). Ported from scripts/build-german-slice.mjs.
+ *
+ *   ORDER ROBUSTNESS (2026-06-06): mixing the two renderings stops a native-only slice from teaching
+ *   German order so well it reads the US/feed-order eval as a "collapse". See
+ *   docs/articles/evals/resolver-geo/2026-06-06-anchor-pilot.md (the order-artifact correction).
+ *
+ *   TWO REGISTERS THE OA TUPLES DO NOT CARRY (#1946). A comma-free single line — `Neusser Str. 12 Nippes
+ *   50733 Köln`, the dictation / one-field-form register — segments as ONE unit at stage 2, and with one
+ *   segment the placetype-pair prior never fires, so the model reads `Nippes` as a second street and the
+ *   one-value-per-tag projection deletes it. `--comma-free-fraction` renders that many native-order rows
+ *   with `" "` as the line separator. And OA rows have no district at all, so `--ortsteil-fraction` rows
+ *   borrow a WOF Ortsteil of the tuple's own locality as `dependent_locality` — 67,532 DE neighbourhoods
+ *   carry a locality ancestor in the admin database, `Köln-Nippes` among them. The German spelling is
+ *   recovered from the `names` table ({@link ortsteilSurface}): WOF's `spr.name` for DE neighbourhoods is
+ *   the ASCII-folded label (`Bocklemuend`), which no German ever types.
+ */
+import { type CorpusRecipe } from "#recipes/scaffold";
+/**
+ * The surface an Ortsteil is written with in an address, from WOF's rows for it.
+ *
+ * `spr.name` for a DE neighbourhood is the ASCII-folded label; the `names` rows in `deu` carry the German spelling
+ * beside unrelated labels for co-located features (`Bocklemuend` → `Bocklemünd`, `Jüdischer Friedhof Bocklemünd`,
+ * `Menara-Garten`). The German name is the one that spells the same label as `spr.name` under either of WOF's ASCII
+ * folds ({@link foldGerman}); with none, `spr.name` stands. WOF also prefixes some Ortsteile with their city
+ * (`Köln-Nippes`), a form no envelope carries once the city is its own line, so a leading `<locality>-` is dropped when
+ * something is left after it.
+ */
+export declare function ortsteilSurface(sprName: string, deuNames: readonly string[], locality: string): string;
+/**
+ * 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 germanRecipe: CorpusRecipe;
+//# sourceMappingURL=locale.d.ts.map

package/out/de/recipes/locale.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"locale.d.ts","sourceRoot":"","sources":["../../../lib/de/recipes/locale.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAUH,OAAO,EAAwB,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAA;AA0D3E;;;;;;;;;GASG;AACH,wBAAgB,eAAe,CAAC,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,SAAS,MAAM,EAAE,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOtG;AAoFD;;;GAGG;AACH,eAAO,MAAM,YAAY,EAAE,YAiK1B,CAAA"}

package/out/de/recipes/locale.js

@@ -0,0 +1,289 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `german` slice recipe — German coverage rows from REAL OpenAddresses tuples (Berlin + Saxony,
+ *   cached zips). Each sampled tuple is rendered via {@link synthesizeGermanRow} in BOTH orders —
+ *   `--intl-fraction` (default 0.4) in international order (house-first / postcode-after-city), the
+ *   rest in idiomatic German order — then aligned to BIO. Generate-mode: it builds a tuple pool
+ *   from the cached zips, then draws `--count` rows from it with the passed `random` (so the emit
+ *   stream matches the legacy reservoir-sample loop). Ported from scripts/build-german-slice.mjs.
+ *
+ *   ORDER ROBUSTNESS (2026-06-06): mixing the two renderings stops a native-only slice from teaching
+ *   German order so well it reads the US/feed-order eval as a "collapse". See
+ *   docs/articles/evals/resolver-geo/2026-06-06-anchor-pilot.md (the order-artifact correction).
+ *
+ *   TWO REGISTERS THE OA TUPLES DO NOT CARRY (#1946). A comma-free single line — `Neusser Str. 12 Nippes
+ *   50733 Köln`, the dictation / one-field-form register — segments as ONE unit at stage 2, and with one
+ *   segment the placetype-pair prior never fires, so the model reads `Nippes` as a second street and the
+ *   one-value-per-tag projection deletes it. `--comma-free-fraction` renders that many native-order rows
+ *   with `" "` as the line separator. And OA rows have no district at all, so `--ortsteil-fraction` rows
+ *   borrow a WOF Ortsteil of the tuple's own locality as `dependent_locality` — 67,532 DE neighbourhoods
+ *   carry a locality ancestor in the admin database, `Köln-Nippes` among them. The German spelling is
+ *   recovered from the `names` table ({@link ortsteilSurface}): WOF's `spr.name` for DE neighbourhoods is
+ *   the ASCII-folded label (`Bocklemuend`), which no German ever types.
+ */
+import { dataRootPath } from "@mailwoman/core/data-root";
+import { pathExists } from "@mailwoman/core/fs/readers";
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
+import { DatabaseClient } from "@mailwoman/sqlite/client";
+import { stableSourceID } from "#adapters/utils";
+import { readZippedCSVRecords } from "#recipes/scaffold";
+import { synthesizeGermanRow } from "#synthesizers/locale";
+import { alignRow } from "#utils";
+/**
+ * `region` is the Bundesland the source covers. OA's REGION column is empty for DE, but the region is implied by the
+ * per-state file — the international order needs it for the "City, Region Postcode" tail (v0.9.3 / #327). berlin.csv →
+ * Berlin (a city-state, region==locality); sn/statewide → Sachsen.
+ */
+const SOURCES = [
+    { zip: dataRootPath("oa-cache", "de__berlin.zip"), csv: "de/berlin.csv", region: "Berlin" },
+    { zip: dataRootPath("oa-cache", "de__sn__statewide.zip"), csv: "de/sn/statewide.csv", region: "Sachsen" },
+];
+/**
+ * The two ASCII spellings WOF uses for one German label, so a `names.deu` row can be matched to the `spr.name` it
+ * spells. WOF folds `Bocklemünd` to `Bocklemuend` in one record and `Schöneberg` to `Schoneberg` in another — the
+ * transliteration (`ö` → `oe`, `ß` → `ss`) and the plain diacritic strip (`ö` → `o`) both occur — so a match is against
+ * either form, lower-cased.
+ */
+function foldGerman(surface) {
+    const lower = surface.toLowerCase();
+    const stripped = lower
+        .normalize("NFD")
+        .replaceAll(/[\u0300-\u036F]/g, "")
+        .replaceAll("ß", "ss");
+    const transliterated = lower
+        .replaceAll("ä", "ae")
+        .replaceAll("ö", "oe")
+        .replaceAll("ü", "ue")
+        .replaceAll("ß", "ss")
+        .normalize("NFD")
+        .replaceAll(/[\u0300-\u036F]/g, "");
+    return [transliterated, stripped];
+}
+/**
+ * Whether two surfaces spell the same German label under either of WOF's folds.
+ */
+function sameGermanLabel(left, right) {
+    const [leftT, leftS] = foldGerman(left);
+    const [rightT, rightS] = foldGerman(right);
+    return leftT === rightT || leftS === rightS || leftT === rightS || leftS === rightT;
+}
+/**
+ * The surface an Ortsteil is written with in an address, from WOF's rows for it.
+ *
+ * `spr.name` for a DE neighbourhood is the ASCII-folded label; the `names` rows in `deu` carry the German spelling
+ * beside unrelated labels for co-located features (`Bocklemuend` → `Bocklemünd`, `Jüdischer Friedhof Bocklemünd`,
+ * `Menara-Garten`). The German name is the one that spells the same label as `spr.name` under either of WOF's ASCII
+ * folds ({@link foldGerman}); with none, `spr.name` stands. WOF also prefixes some Ortsteile with their city
+ * (`Köln-Nippes`), a form no envelope carries once the city is its own line, so a leading `<locality>-` is dropped when
+ * something is left after it.
+ */
+export function ortsteilSurface(sprName, deuNames, locality) {
+    const german = deuNames.find((name) => sameGermanLabel(name, sprName)) ?? sprName;
+    const prefix = `${locality}-`;
+    return german.toLowerCase().startsWith(prefix.toLowerCase()) && german.length > prefix.length
+        ? german.slice(prefix.length)
+        : german;
+}
+/**
+ * Every current DE neighbourhood with a locality ancestor, keyed by the folded locality name → Ortsteil surfaces. Empty
+ * when the admin database is not readable; the recipe then emits no Ortsteil rows and says so, rather than failing a
+ * build over an optional register.
+ */
+async function readOrtsteilPool(adminDB) {
+    const pool = new Map();
+    if (!(await pathExists(adminDB)))
+        return pool;
+    using db = new DatabaseClient(adminDB, { readOnly: true });
+    const rows = await db
+        .selectFrom("spr as s")
+        .innerJoin("ancestors as a", "a.id", "s.id")
+        .innerJoin("spr as p", "p.id", "a.ancestor_id")
+        .select(["s.id as id", "s.name as name", "p.name as locality"])
+        .where("s.country", "=", "DE")
+        .where("s.placetype", "=", "neighbourhood")
+        .where("s.is_current", "=", 1)
+        .where("s.is_deprecated", "=", 0)
+        .where("a.ancestor_placetype", "=", "locality")
+        .execute();
+    const deuNames = new Map();
+    for (const row of await db
+        .selectFrom("names")
+        .select(["id", "name"])
+        .where("country", "=", "DE")
+        .where("placetype", "=", "neighbourhood")
+        .where("language", "=", "deu")
+        .execute()) {
+        const list = deuNames.get(row.id) ?? [];
+        list.push(row.name);
+        deuNames.set(row.id, list);
+    }
+    for (const row of rows) {
+        if (!row.name || !row.locality)
+            continue;
+        const surface = ortsteilSurface(row.name, deuNames.get(row.id) ?? [], row.locality);
+        const key = row.locality.toLowerCase();
+        const list = pool.get(key) ?? [];
+        if (!list.includes(surface)) {
+            list.push(surface);
+        }
+        pool.set(key, list);
+    }
+    return pool;
+}
+/**
+ * Stream real German tuples out of a cached OA zip.
+ */
+async function readGermanTuples(source) {
+    const tuples = [];
+    const seen = new Set();
+    for await (const row of readZippedCSVRecords(source.zip, source.csv)) {
+        const street = row.street ?? "";
+        const locality = row.city ?? "";
+        if (!street || !locality)
+            continue;
+        const house_number = row.number ?? "";
+        const postcode = row.postcode ?? "";
+        // OA's REGION column is empty for DE — fall back to the source's Bundesland (set per file).
+        const region = row.region || source.region || "";
+        const key = `${house_number}|${street}|${locality}|${postcode}`.toLowerCase();
+        if (seen.has(key))
+            continue;
+        seen.add(key);
+        tuples.push({ house_number, street, locality, region, postcode });
+    }
+    return tuples;
+}
+/**
+ * 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 germanRecipe = {
+    name: "german",
+    description: "German coverage rows from real OA tuples (Berlin/Saxony), both orders → synthesizeGermanRow",
+    mode: "generate",
+    options: [
+        { flag: "--intl-fraction <f>", description: "Fraction rendered international order. Default 0.4" },
+        {
+            flag: "--comma-free-fraction <f>",
+            description: "Fraction of native-order rows rendered with no commas (the single-line register, #1946). Default 0.3",
+        },
+        {
+            flag: "--ortsteil-fraction <f>",
+            description: "Fraction of rows carrying a WOF Ortsteil of the locality as dependent_locality. Default 0.3",
+        },
+        {
+            flag: "--admin-db <path>",
+            description: "WOF admin database for the Ortsteil pool. Default $MAILWOMAN_DATA_ROOT/wof/admin-global-priority-importance.db",
+        },
+    ],
+    async run(opts, write) {
+        // Emit PRNG: the legacy build-german-slice.mjs seeded mulberry32(opts.seed).
+        const random = makeMulberry32(opts.seed);
+        const source = opts.sourceName ?? "synth-german";
+        const intlFraction = opts.intlFraction ?? 0.4;
+        const commaFreeFraction = opts.commaFreeFraction ?? 0.3;
+        const ortsteilFraction = opts.ortsteilFraction ?? 0.3;
+        for (const [flag, value] of [
+            ["--intl-fraction", intlFraction],
+            ["--comma-free-fraction", commaFreeFraction],
+            ["--ortsteil-fraction", ortsteilFraction],
+        ]) {
+            if (!(value >= 0 && value <= 1)) {
+                throw new Error(`${flag} must be in [0, 1], got ${value}`);
+            }
+        }
+        const adminDB = opts.adminDB ?? String(dataRootPath("wof", "admin-global-priority-importance.db"));
+        const ortsteile = ortsteilFraction > 0 ? await readOrtsteilPool(adminDB) : new Map();
+        if (ortsteilFraction > 0) {
+            console.error(ortsteile.size
+                ? `  Ortsteil pool: ${[...ortsteile.values()].reduce((n, list) => n + list.length, 0)} surfaces under ${ortsteile.size} localities (${adminDB})`
+                : `  Ortsteil pool: none — ${adminDB} is not readable, so no row carries a dependent_locality`);
+        }
+        const count = opts.count ?? 4000;
+        // Pool real tuples from every German source, then sample `count` rows from it.
+        const pool = [];
+        for (const s of SOURCES) {
+            const t = await readGermanTuples(s);
+            console.error(`  ${s.csv}: ${t.length} unique tuples`);
+            for (const x of t) {
+                pool.push(x);
+            } // NOT pool.push(...t) — spreading ~840K args overflows the stack
+        }
+        if (!pool.length) {
+            throw new Error(`No German tuples found — are the cached zips present in ${dataRootPath("oa-cache")}?`);
+        }
+        let emitted = 0;
+        let skipped = 0;
+        let guard = 0;
+        let withOrtsteil = 0;
+        let commaFree = 0;
+        const N = pool.length;
+        while (emitted < count && guard++ < count * 6) {
+            const drawn = pool[Math.floor(random() * N)];
+            // Per-row order: `--intl-fraction` of rows render house-first / postcode-after-city (the US/feed
+            // layout), the rest in idiomatic German order. Same components either way.
+            const order = random() < intlFraction ? "international" : "native";
+            // The two registers OA never wrote (#1946), each drawn independently of the order so every combination
+            // occurs: an Ortsteil borrowed from the tuple's own locality, and a native line with no commas.
+            const localOrtsteile = ortsteile.get(drawn.locality.toLowerCase());
+            const ortsteil = localOrtsteile && random() < ortsteilFraction
+                ? localOrtsteile[Math.floor(random() * localOrtsteile.length)]
+                : undefined;
+            const base = ortsteil ? { ...drawn, dependent_locality: ortsteil } : drawn;
+            const separator = order === "native" && random() < commaFreeFraction ? " " : ", ";
+            const synth = synthesizeGermanRow(base, { random, order, separator });
+            if (!synth) {
+                skipped++;
+                continue;
+            }
+            // --golden: emit per-locale-f1 eval rows ({raw, components}) instead of aligned BIO. `order`
+            // rides along so the eval can stratify native vs international.
+            if (opts.golden) {
+                write(JSON.stringify({ raw: synth.raw, components: synth.components, country: "DE", order }) + "\n");
+                emitted++;
+                continue;
+            }
+            const sourceID = stableSourceID(source, {
+                street: synth.components.street,
+                house_number: synth.components.house_number,
+                locality: synth.components.locality,
+                postcode: synth.components.postcode,
+            });
+            const canonical = {
+                raw: synth.raw,
+                components: synth.components,
+                country: "DE",
+                locale: synth.locale,
+                source,
+                source_id: sourceID,
+                corpus_version: "0.4.0",
+                license: `OpenAddresses DE (Berlin/Saxony) tuples, rendered ${order}-order — see ingest SOURCES`,
+            };
+            const aligned = alignRow(canonical);
+            if (aligned.kind !== "labeled" || !aligned.row) {
+                skipped++;
+                continue;
+            }
+            write(JSON.stringify({
+                ...aligned.row,
+                synth_method: "german",
+                synth_order: order,
+                synth_separator: separator === " " ? "space" : "comma",
+                synth_base_id: null,
+            }) + "\n");
+            emitted++;
+            if (ortsteil) {
+                withOrtsteil++;
+            }
+            if (separator === " ") {
+                commaFree++;
+            }
+        }
+        console.error(`  emitted ${emitted}: ${withOrtsteil} with an Ortsteil, ${commaFree} comma-free`);
+        return { emitted, skipped };
+    },
+};
+//# sourceMappingURL=locale.js.map

package/out/de/recipes/locale.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"locale.js","sourceRoot":"","sources":["../../../lib/de/recipes/locale.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAA;AACxD,OAAO,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAA;AACvD,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AAEpE,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAA;AAGzD,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAA;AAChD,OAAO,EAAE,oBAAoB,EAAqB,MAAM,mBAAmB,CAAA;AAC3E,OAAO,EAAE,mBAAmB,EAAwB,MAAM,sBAAsB,CAAA;AAChF,OAAO,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAA;AAWjC;;;;GAIG;AACH,MAAM,OAAO,GAAmB;IAC/B,EAAE,GAAG,EAAE,YAAY,CAAC,UAAU,EAAE,gBAAgB,CAAC,EAAE,GAAG,EAAE,eAAe,EAAE,MAAM,EAAE,QAAQ,EAAE;IAC3F,EAAE,GAAG,EAAE,YAAY,CAAC,UAAU,EAAE,uBAAuB,CAAC,EAAE,GAAG,EAAE,qBAAqB,EAAE,MAAM,EAAE,SAAS,EAAE;CACzG,CAAA;AAED;;;;;GAKG;AACH,SAAS,UAAU,CAAC,OAAe;IAClC,MAAM,KAAK,GAAG,OAAO,CAAC,WAAW,EAAE,CAAA;IAEnC,MAAM,QAAQ,GAAG,KAAK;SACpB,SAAS,CAAC,KAAK,CAAC;SAChB,UAAU,CAAC,kBAAkB,EAAE,EAAE,CAAC;SAClC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAA;IAEvB,MAAM,cAAc,GAAG,KAAK;SAC1B,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC;SACrB,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC;SACrB,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC;SACrB,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC;SACrB,SAAS,CAAC,KAAK,CAAC;SAChB,UAAU,CAAC,kBAAkB,EAAE,EAAE,CAAC,CAAA;IAEpC,OAAO,CAAC,cAAc,EAAE,QAAQ,CAAC,CAAA;AAClC,CAAC;AAED;;GAEG;AACH,SAAS,eAAe,CAAC,IAAY,EAAE,KAAa;IACnD,MAAM,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,UAAU,CAAC,IAAI,CAAC,CAAA;IACvC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,UAAU,CAAC,KAAK,CAAC,CAAA;IAE1C,OAAO,KAAK,KAAK,MAAM,IAAI,KAAK,KAAK,MAAM,IAAI,KAAK,KAAK,MAAM,IAAI,KAAK,KAAK,MAAM,CAAA;AACpF,CAAC;AAED;;;;;;;;;GASG;AACH,MAAM,UAAU,eAAe,CAAC,OAAe,EAAE,QAA2B,EAAE,QAAgB;IAC7F,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,eAAe,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,IAAI,OAAO,CAAA;IACjF,MAAM,MAAM,GAAG,GAAG,QAAQ,GAAG,CAAA;IAE7B,OAAO,MAAM,CAAC,WAAW,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,IAAI,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM;QAC5F,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC;QAC7B,CAAC,CAAC,MAAM,CAAA;AACV,CAAC;AAED;;;;GAIG;AACH,KAAK,UAAU,gBAAgB,CAAC,OAAe;IAC9C,MAAM,IAAI,GAAG,IAAI,GAAG,EAAoB,CAAA;IAExC,IAAI,CAAC,CAAC,MAAM,UAAU,CAAC,OAAO,CAAC,CAAC;QAAE,OAAO,IAAI,CAAA;IAE7C,MAAM,EAAE,GAAG,IAAI,cAAc,CAAc,OAAO,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAA;IAEvE,MAAM,IAAI,GAAG,MAAM,EAAE;SACnB,UAAU,CAAC,UAAU,CAAC;SACtB,SAAS,CAAC,gBAAgB,EAAE,MAAM,EAAE,MAAM,CAAC;SAC3C,SAAS,CAAC,UAAU,EAAE,MAAM,EAAE,eAAe,CAAC;SAC9C,MAAM,CAAC,CAAC,YAAY,EAAE,gBAAgB,EAAE,oBAAoB,CAAC,CAAC;SAC9D,KAAK,CAAC,WAAW,EAAE,GAAG,EAAE,IAAI,CAAC;SAC7B,KAAK,CAAC,aAAa,EAAE,GAAG,EAAE,eAAe,CAAC;SAC1C,KAAK,CAAC,cAAc,EAAE,GAAG,EAAE,CAAC,CAAC;SAC7B,KAAK,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,CAAC;SAChC,KAAK,CAAC,sBAAsB,EAAE,GAAG,EAAE,UAAU,CAAC;SAC9C,OAAO,EAAE,CAAA;IAEX,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAoB,CAAA;IAE5C,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE;SACxB,UAAU,CAAC,OAAO,CAAC;SACnB,MAAM,CAAC,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;SACtB,KAAK,CAAC,SAAS,EAAE,GAAG,EAAE,IAAI,CAAC;SAC3B,KAAK,CAAC,WAAW,EAAE,GAAG,EAAE,eAAe,CAAC;SACxC,KAAK,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,CAAC;SAC7B,OAAO,EAAE,EAAE,CAAC;QACb,MAAM,IAAI,GAAG,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,CAAA;QAEvC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;QACnB,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,CAAA;IAC3B,CAAC;IAED,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACxB,IAAI,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC,GAAG,CAAC,QAAQ;YAAE,SAAQ;QAExC,MAAM,OAAO,GAAG,eAAe,CAAC,GAAG,CAAC,IAAI,EAAE,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAA;QACnF,MAAM,GAAG,GAAG,GAAG,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAA;QACtC,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,CAAA;QAEhC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;YAC7B,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAA;QACnB,CAAC;QAED,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAA;IACpB,CAAC;IAED,OAAO,IAAI,CAAA;AACZ,CAAC;AAED;;GAEG;AACH,KAAK,UAAU,gBAAgB,CAAC,MAAoB;IACnD,MAAM,MAAM,GAAsB,EAAE,CAAA;IACpC,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAE9B,IAAI,KAAK,EAAE,MAAM,GAAG,IAAI,oBAAoB,CAAC,MAAM,CAAC,GAAG,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC;QACtE,MAAM,MAAM,GAAG,GAAG,CAAC,MAAM,IAAI,EAAE,CAAA;QAC/B,MAAM,QAAQ,GAAG,GAAG,CAAC,IAAI,IAAI,EAAE,CAAA;QAE/B,IAAI,CAAC,MAAM,IAAI,CAAC,QAAQ;YAAE,SAAQ;QAClC,MAAM,YAAY,GAAG,GAAG,CAAC,MAAM,IAAI,EAAE,CAAA;QACrC,MAAM,QAAQ,GAAG,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAA;QACnC,4FAA4F;QAC5F,MAAM,MAAM,GAAG,GAAG,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,IAAI,EAAE,CAAA;QAChD,MAAM,GAAG,GAAG,GAAG,YAAY,IAAI,MAAM,IAAI,QAAQ,IAAI,QAAQ,EAAE,CAAC,WAAW,EAAE,CAAA;QAE7E,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;YAAE,SAAQ;QAC3B,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;QACb,MAAM,CAAC,IAAI,CAAC,EAAE,YAAY,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC,CAAA;IAClE,CAAC;IAED,OAAO,MAAM,CAAA;AACd,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,YAAY,GAAiB;IACzC,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,6FAA6F;IAC1G,IAAI,EAAE,UAAU;IAChB,OAAO,EAAE;QACR,EAAE,IAAI,EAAE,qBAAqB,EAAE,WAAW,EAAE,oDAAoD,EAAE;QAClG;YACC,IAAI,EAAE,2BAA2B;YACjC,WAAW,EACV,sGAAsG;SACvG;QACD;YACC,IAAI,EAAE,yBAAyB;YAC/B,WAAW,EAAE,6FAA6F;SAC1G;QACD;YACC,IAAI,EAAE,mBAAmB;YACzB,WAAW,EACV,gHAAgH;SACjH;KACD;IACD,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,6EAA6E;QAC7E,MAAM,MAAM,GAAG,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACxC,MAAM,MAAM,GAAG,IAAI,CAAC,UAAU,IAAI,cAAc,CAAA;QAChD,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,GAAG,CAAA;QAC7C,MAAM,iBAAiB,GAAG,IAAI,CAAC,iBAAiB,IAAI,GAAG,CAAA;QACvD,MAAM,gBAAgB,GAAG,IAAI,CAAC,gBAAgB,IAAI,GAAG,CAAA;QAErD,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI;YAC3B,CAAC,iBAAiB,EAAE,YAAY,CAAC;YACjC,CAAC,uBAAuB,EAAE,iBAAiB,CAAC;YAC5C,CAAC,qBAAqB,EAAE,gBAAgB,CAAC;SAChC,EAAE,CAAC;YACZ,IAAI,CAAC,CAAC,KAAK,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,EAAE,CAAC;gBACjC,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,2BAA2B,KAAK,EAAE,CAAC,CAAA;YAC3D,CAAC;QACF,CAAC;QAED,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,MAAM,CAAC,YAAY,CAAC,KAAK,EAAE,qCAAqC,CAAC,CAAC,CAAA;QAClG,MAAM,SAAS,GAAG,gBAAgB,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAoB,CAAA;QAEtG,IAAI,gBAAgB,GAAG,CAAC,EAAE,CAAC;YAC1B,OAAO,CAAC,KAAK,CACZ,SAAS,CAAC,IAAI;gBACb,CAAC,CAAC,oBAAoB,CAAC,GAAG,SAAS,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC,mBAAmB,SAAS,CAAC,IAAI,gBAAgB,OAAO,GAAG;gBAChJ,CAAC,CAAC,2BAA2B,OAAO,0DAA0D,CAC/F,CAAA;QACF,CAAC;QAED,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,IAAI,CAAA;QAEhC,+EAA+E;QAC/E,MAAM,IAAI,GAAsB,EAAE,CAAA;QAElC,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;YACzB,MAAM,CAAC,GAAG,MAAM,gBAAgB,CAAC,CAAC,CAAC,CAAA;YAEnC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,MAAM,gBAAgB,CAAC,CAAA;YAEtD,KAAK,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC;gBACnB,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;YACb,CAAC,CAAC,iEAAiE;QACpE,CAAC;QAED,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;YAClB,MAAM,IAAI,KAAK,CAAC,2DAA2D,YAAY,CAAC,UAAU,CAAC,GAAG,CAAC,CAAA;QACxG,CAAC;QAED,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,KAAK,GAAG,CAAC,CAAA;QACb,IAAI,YAAY,GAAG,CAAC,CAAA;QACpB,IAAI,SAAS,GAAG,CAAC,CAAA;QACjB,MAAM,CAAC,GAAG,IAAI,CAAC,MAAM,CAAA;QAErB,OAAO,OAAO,GAAG,KAAK,IAAI,KAAK,EAAE,GAAG,KAAK,GAAG,CAAC,EAAE,CAAC;YAC/C,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAE,CAAA;YAC7C,iGAAiG;YACjG,2EAA2E;YAC3E,MAAM,KAAK,GAAG,MAAM,EAAE,GAAG,YAAY,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,QAAQ,CAAA;YAClE,uGAAuG;YACvG,gGAAgG;YAChG,MAAM,cAAc,GAAG,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,CAAA;YAElE,MAAM,QAAQ,GACb,cAAc,IAAI,MAAM,EAAE,GAAG,gBAAgB;gBAC5C,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;gBAC9D,CAAC,CAAC,SAAS,CAAA;YAEb,MAAM,IAAI,GAAoB,QAAQ,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,EAAE,kBAAkB,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,KAAK,CAAA;YAC3F,MAAM,SAAS,GAAG,KAAK,KAAK,QAAQ,IAAI,MAAM,EAAE,GAAG,iBAAiB,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAA;YACjF,MAAM,KAAK,GAAG,mBAAmB,CAAC,IAAI,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC,CAAA;YAErE,IAAI,CAAC,KAAK,EAAE,CAAC;gBACZ,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,6FAA6F;YAC7F,gEAAgE;YAChE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;gBACjB,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,UAAU,EAAE,KAAK,CAAC,UAAU,EAAE,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,GAAG,IAAI,CAAC,CAAA;gBAEpG,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,MAAM,QAAQ,GAAG,cAAc,CAAC,MAAM,EAAE;gBACvC,MAAM,EAAE,KAAK,CAAC,UAAU,CAAC,MAAM;gBAC/B,YAAY,EAAE,KAAK,CAAC,UAAU,CAAC,YAAY;gBAC3C,QAAQ,EAAE,KAAK,CAAC,UAAU,CAAC,QAAQ;gBACnC,QAAQ,EAAE,KAAK,CAAC,UAAU,CAAC,QAAQ;aACnC,CAAC,CAAA;YAEF,MAAM,SAAS,GAAG;gBACjB,GAAG,EAAE,KAAK,CAAC,GAAG;gBACd,UAAU,EAAE,KAAK,CAAC,UAAU;gBAC5B,OAAO,EAAE,IAAI;gBACb,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,MAAM;gBACN,SAAS,EAAE,QAAQ;gBACnB,cAAc,EAAE,OAAO;gBACvB,OAAO,EAAE,qDAAqD,KAAK,6BAA6B;aAChG,CAAA;YAED,MAAM,OAAO,GAAG,QAAQ,CAAC,SAA2C,CAAC,CAAA;YAErE,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC;gBAChD,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,KAAK,CACJ,IAAI,CAAC,SAAS,CAAC;gBACd,GAAG,OAAO,CAAC,GAAG;gBACd,YAAY,EAAE,QAAQ;gBACtB,WAAW,EAAE,KAAK;gBAClB,eAAe,EAAE,SAAS,KAAK,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO;gBACtD,aAAa,EAAE,IAAI;aACnB,CAAC,GAAG,IAAI,CACT,CAAA;YAED,OAAO,EAAE,CAAA;YAET,IAAI,QAAQ,EAAE,CAAC;gBACd,YAAY,EAAE,CAAA;YACf,CAAC;YAED,IAAI,SAAS,KAAK,GAAG,EAAE,CAAC;gBACvB,SAAS,EAAE,CAAA;YACZ,CAAC;QACF,CAAC;QAED,OAAO,CAAC,KAAK,CAAC,aAAa,OAAO,KAAK,YAAY,sBAAsB,SAAS,aAAa,CAAC,CAAA;QAEhG,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAC5B,CAAC;CACD,CAAA"}

package/out/env.d.ts

@@ -28,10 +28,10 @@ export declare const $public: Record<never, never> & {
     MAILWOMAN_CONFIG_ROOT: string;
     MAILWOMAN_CACHE_ROOT: string;
     MAILWOMAN_TEMP_ROOT: string;
+    MAILWOMAN_COARSE_PLACER_DIR?: string | undefined;
     MAILWOMAN_DOCS_URL?: string | undefined;
     MAILWOMAN_LICENSE_KEY?: string | undefined;
     MAILWOMAN_LICENSE_URL?: string | undefined;
-    MAILWOMAN_COARSE_PLACER_DIR?: string | undefined;
 } & {
     MAILWOMAN_RESUME?: string | undefined;
 };

package/out/fr/adapters/ban/adapter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"adapter.d.ts","sourceRoot":"","sources":["../../../../lib/fr/adapters/ban/adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA+BG;AAOH,OAAO,KAAK,EAAgC,aAAa,EAAE,MAAM,QAAQ,CAAA;AAEzE;;;GAGG;AACH,eAAO,MAAM,cAAc,QAAQ,CAAA;AA4BnC,wBAAgB,gBAAgB,IAAI,aAAa,CAiFhD;AAED;;GAEG;AACH,eAAO,MAAM,UAAU,eAAqB,CAAA"}

package/out/fr/adapters/ban/adapter.js

@@ -0,0 +1,123 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `ban`: Base Adresse Nationale CSV adapter (FR street-level).
+ *
+ *   Input: a CSV dump from `adresse.data.gouv.fr` (semicolon-separated, ~25M rows nationally). The
+ *   adapter only reads the small set of columns needed for the corpus:
+ *
+ *   - `numero` → `house_number`
+ *   - `rep` → repetition index ("bis", "ter") appended to house_number
+ *   - `nom_voie` → `street` (full road name; includes the prefix "Rue", "Avenue", etc.)
+ *   - `code_postal` → `postcode`
+ *   - `nom_commune` → `locality`
+ *
+ *   `region` and `country` are not in BAN. The adapter stamps `country: "FR"` on every row; region is
+ *   left for the wof-postalcode + wof-admin cross-reference at corpus build time (a future pass;
+ *   for Phase 1 the row's region is simply absent).
+ *
+ *   License: the official BAN (adresse.data.gouv.fr) is DUAL-licensed — Licence Ouverte 2.0 (Etalab,
+ *   attribution-only) OR ODbL (share-alike). We ELECT Licence Ouverte 2.0 (issue #26 Tier B:
+ *   allowed for training with attribution; the ODbL option's share-alike obligation would defeat
+ *   the proprietary-weights goal). Stamped onto every row as `Licence Ouverte 2.0` — NOT the older
+ *   conservative `ODbL-1.0` label, which wrongly read as Tier-C-denied in the corpus license audit.
+ *   The model card MUST carry the BAN attribution (Tier B obligation).
+ *
+ *   The adapter is streaming-aware: `CSVSpliterator.fromAsync` reads the `;`-delimited dump row by
+ *   row, so a 25M-row file never sits in memory. Honors `opts.limit` for fixture / smoke runs,
+ *   `opts.signal` for cancellation, and `opts.country` for a self-consistency check (errors if
+ *   country !== FR).
+ */
+import { formatAddressRow } from "@mailwoman/codex/address-format";
+import { CSVSpliterator } from "spliterator";
+import { stableSourceID } from "#adapters/utils";
+import { decomposeFrStreet } from "#fr/adapters/ban/street-decompose";
+/**
+ * Registry id for this adapter. Stamped into every row it emits, so a corpus record can be traced back to the dataset
+ * it came from.
+ */
+export const BAN_ADAPTER_ID = "ban";
+/**
+ * Compose `house_number` from `numero` + `rep`. BAN uses `rep` for repetition indices ("bis", "ter", "quater") that
+ * follow the house number. Result: `"10 bis"`, `"45"`, etc.
+ */
+function composeHouseNumber(numero, rep) {
+    const n = numero.trim();
+    const r = rep.trim();
+    if (!n)
+        return "";
+    return r ? `${n} ${r}` : n;
+}
+export function createBanAdapter() {
+    return {
+        id: BAN_ADAPTER_ID,
+        defaultLicense: "Licence Ouverte 2.0",
+        description: "Base Adresse Nationale (FR): house-number-level street addresses (~25M rows).",
+        async *rows(opts) {
+            if (opts.country && opts.country !== "FR") {
+                throw new Error(`ban adapter: only FR supported, got country=${opts.country}`);
+            }
+            const rows = CSVSpliterator.fromAsync(opts.inputPath, {
+                normalizeKeys: false,
+                columnDelimiter: ";",
+            });
+            let emitted = 0;
+            for await (const record of rows) {
+                if (opts.signal?.aborted)
+                    break;
+                if (opts.limit !== undefined && emitted >= opts.limit)
+                    break;
+                const house = composeHouseNumber(record.numero ?? "", record.rep ?? "");
+                const street = (record.nom_voie ?? "").trim();
+                const postcode = (record.code_postal ?? "").trim();
+                const locality = (record.nom_commune ?? "").trim();
+                if (!street || !locality)
+                    continue;
+                if (!house && !postcode)
+                    continue;
+                const decomposed = decomposeFrStreet(street);
+                const components = {};
+                if (house) {
+                    components.house_number = house;
+                }
+                if (decomposed.prefix) {
+                    components.street_prefix = decomposed.prefix;
+                }
+                if (decomposed.street) {
+                    components.street = decomposed.street;
+                }
+                if (postcode) {
+                    components.postcode = postcode;
+                }
+                if (locality) {
+                    components.locality = locality;
+                }
+                const rendered = formatAddressRow(components, "FR", { singleLine: true });
+                if (!rendered)
+                    continue;
+                const { raw, components: aligned } = rendered;
+                const sourceID = record.id?.trim()
+                    ? `${BAN_ADAPTER_ID}-${record.id.trim()}`
+                    : stableSourceID(BAN_ADAPTER_ID, aligned);
+                yield {
+                    raw,
+                    components: aligned,
+                    country: "FR",
+                    locale: "fr-FR",
+                    source: BAN_ADAPTER_ID,
+                    source_id: sourceID,
+                    corpus_version: "",
+                    license: "Licence Ouverte 2.0",
+                };
+                emitted++;
+            }
+        },
+    };
+}
+/**
+ * The configured adapter instance registered with the corpus builder.
+ */
+export const banAdapter = createBanAdapter();
+//# sourceMappingURL=adapter.js.map

package/out/fr/adapters/ban/adapter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"adapter.js","sourceRoot":"","sources":["../../../../lib/fr/adapters/ban/adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA+BG;AAEH,OAAO,EAAE,gBAAgB,EAAE,MAAM,iCAAiC,CAAA;AAClE,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAA;AAE5C,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAA;AAChD,OAAO,EAAE,iBAAiB,EAAE,MAAM,mCAAmC,CAAA;AAGrE;;;GAGG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG,KAAK,CAAA;AAenC;;;GAGG;AACH,SAAS,kBAAkB,CAAC,MAAc,EAAE,GAAW;IACtD,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAA;IACvB,MAAM,CAAC,GAAG,GAAG,CAAC,IAAI,EAAE,CAAA;IAEpB,IAAI,CAAC,CAAC;QAAE,OAAO,EAAE,CAAA;IAEjB,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAA;AAC3B,CAAC;AAED,MAAM,UAAU,gBAAgB;IAC/B,OAAO;QACN,EAAE,EAAE,cAAc;QAClB,cAAc,EAAE,qBAAqB;QACrC,WAAW,EAAE,+EAA+E;QAE5F,KAAK,CAAC,CAAC,IAAI,CAAC,IAAoB;YAC/B,IAAI,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;gBAC3C,MAAM,IAAI,KAAK,CAAC,+CAA+C,IAAI,CAAC,OAAO,EAAE,CAAC,CAAA;YAC/E,CAAC;YAED,MAAM,IAAI,GAAG,cAAc,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS,EAAE;gBACrD,aAAa,EAAE,KAAK;gBACpB,eAAe,EAAE,GAAG;aACpB,CAAC,CAAA;YAEF,IAAI,OAAO,GAAG,CAAC,CAAA;YAEf,IAAI,KAAK,EAAE,MAAM,MAAM,IAAI,IAA6B,EAAE,CAAC;gBAC1D,IAAI,IAAI,CAAC,MAAM,EAAE,OAAO;oBAAE,MAAK;gBAE/B,IAAI,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,OAAO,IAAI,IAAI,CAAC,KAAK;oBAAE,MAAK;gBAE5D,MAAM,KAAK,GAAG,kBAAkB,CAAC,MAAM,CAAC,MAAM,IAAI,EAAE,EAAE,MAAM,CAAC,GAAG,IAAI,EAAE,CAAC,CAAA;gBACvE,MAAM,MAAM,GAAG,CAAC,MAAM,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;gBAC7C,MAAM,QAAQ,GAAG,CAAC,MAAM,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;gBAClD,MAAM,QAAQ,GAAG,CAAC,MAAM,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAA;gBAElD,IAAI,CAAC,MAAM,IAAI,CAAC,QAAQ;oBAAE,SAAQ;gBAElC,IAAI,CAAC,KAAK,IAAI,CAAC,QAAQ;oBAAE,SAAQ;gBAEjC,MAAM,UAAU,GAAG,iBAAiB,CAAC,MAAM,CAAC,CAAA;gBAE5C,MAAM,UAAU,GAA+B,EAAE,CAAA;gBAEjD,IAAI,KAAK,EAAE,CAAC;oBACX,UAAU,CAAC,YAAY,GAAG,KAAK,CAAA;gBAChC,CAAC;gBAED,IAAI,UAAU,CAAC,MAAM,EAAE,CAAC;oBACvB,UAAU,CAAC,aAAa,GAAG,UAAU,CAAC,MAAM,CAAA;gBAC7C,CAAC;gBAED,IAAI,UAAU,CAAC,MAAM,EAAE,CAAC;oBACvB,UAAU,CAAC,MAAM,GAAG,UAAU,CAAC,MAAM,CAAA;gBACtC,CAAC;gBAED,IAAI,QAAQ,EAAE,CAAC;oBACd,UAAU,CAAC,QAAQ,GAAG,QAAQ,CAAA;gBAC/B,CAAC;gBAED,IAAI,QAAQ,EAAE,CAAC;oBACd,UAAU,CAAC,QAAQ,GAAG,QAAQ,CAAA;gBAC/B,CAAC;gBAED,MAAM,QAAQ,GAAG,gBAAgB,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC,CAAA;gBAEzE,IAAI,CAAC,QAAQ;oBAAE,SAAQ;gBAEvB,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,QAAQ,CAAA;gBAE7C,MAAM,QAAQ,GAAG,MAAM,CAAC,EAAE,EAAE,IAAI,EAAE;oBACjC,CAAC,CAAC,GAAG,cAAc,IAAI,MAAM,CAAC,EAAE,CAAC,IAAI,EAAE,EAAE;oBACzC,CAAC,CAAC,cAAc,CAAC,cAAc,EAAE,OAAO,CAAC,CAAA;gBAE1C,MAAM;oBACL,GAAG;oBACH,UAAU,EAAE,OAAO;oBACnB,OAAO,EAAE,IAAI;oBACb,MAAM,EAAE,OAAO;oBACf,MAAM,EAAE,cAAc;oBACtB,SAAS,EAAE,QAAQ;oBACnB,cAAc,EAAE,EAAE;oBAClB,OAAO,EAAE,qBAAqB;iBAC9B,CAAA;gBAED,OAAO,EAAE,CAAA;YACV,CAAC;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,EAAE,CAAA"}

package/out/fr/adapters/ban/street-decompose.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"street-decompose.d.ts","sourceRoot":"","sources":["../../../../lib/fr/adapters/ban/street-decompose.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAWH,MAAM,WAAW,kBAAkB;IAClC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;IACrB,MAAM,EAAE,MAAM,CAAA;CACd;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,MAAM,GAAG,kBAAkB,CA4BtE"}

package/out/fr/adapters/ban/street-decompose.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"street-decompose.js","sourceRoot":"","sources":["../../../../lib/fr/adapters/ban/street-decompose.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,uBAAuB,EAAE,MAAM,iBAAiB,CAAA;AAEzD;;GAEG;AACH,MAAM,wBAAwB,GAAG,CAAC,CAAA;AAElC,MAAM,eAAe,GAAG,MAAM,uBAAuB,CAAC,IAAI,EAAE,kBAAkB,CAAC,CAAA;AAO/E;;;;;GAKG;AACH,MAAM,UAAU,iBAAiB,CAAC,QAAgB;IACjD,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAA;IAE/B,IAAI,CAAC,OAAO;QAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,CAAA;IAEjD,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAA;IAEnC,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,CAAA;IAE/D,MAAM,IAAI,GAAG,CAAC,CAAS,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAA;IAEjE,iDAAiD;IACjD,IAAI,MAAM,CAAC,MAAM,IAAI,wBAAwB,EAAE,CAAC;QAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,GAAG,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,CAAA;QAEzD,IAAI,eAAe,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;YAClC,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAA;QACnF,CAAC;IACF,CAAC;IAED,yBAAyB;IACzB,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,CAAA;IAE9B,IAAI,eAAe,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;QAChC,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC,CAAE,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAA;IACjE,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,CAAA;AACzC,CAAC"}