@mailwoman/corpus 9.1.0 → 9.3.0

package/out/recipes/fr-order.js

@@ -0,0 +1,197 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `fr-order` slice recipe — French reversed-order coverage (#560). Reads REAL OpenAddresses FR
+ *   tuples (`fr/countrywide.csv` from the cached zip), then for each row picks — by
+ *   `--reversed-fraction` (default 0.5) — whether to render in CANONICAL French order
+ *   (number-street, postcode-city) or one of four REVERSED / postcode-first variants the
+ *   v4.4.0→v0.5.0 regression exposed (the model misses house_number in every reversed one):
+ *
+ *   - A: "47110 Sainte-Livrade-sur-Lot, 69 Allée du Bugatel" (postcode city, HN street)
+ *   - B: "Sainte-Livrade-sur-Lot, 47110, 619 Impasse de la Rose" (city, postcode, HN street)
+ *   - C: "Sainte-Livrade-sur-Lot 59 bis Rue des Ecuries 47110" (city HN street postcode — NO commas)
+ *   - D: "47110, 6 rue de la république, Sainte-Livrade-sur-Lot" (postcode, HN street, city)
+ *
+ *   Sub-modes ride alongside order: `bis`/`ter`/`quater` ordinal suffixes in house_number, and
+ *   ALL-CAPS locality. `--golden` emits a held-out reversed-order eval slice with a different
+ *   seed.
+ *
+ *   The inline synthesis (the OA-CSV reader, the ordinal/all-caps tables, the canonical + reversed
+ *   renderers) is ported faithfully from scripts/build-fr-order-slice.mjs. This is a
+ *   `generate`-mode recipe that still reads REAL tuples off disk — `--count` bounds the OUTPUT, not
+ *   the input. The passed `random` (the framework LCG) is consumed in the exact call order the
+ *   legacy script used.
+ *
+ *   NOT ported (diagnostic-only, no effect on emitted bytes): the post-run `runSpanCheck` self-check
+ *   (it reads the finished file back with a separate PRNG and prints to stderr; the recipe's output
+ *   stream is still open during `run`), and the dead `renderReversed` helper (the legacy `main`
+ *   inlined the variant logic and never called it).
+ */
+import { dataRootPath } from "@mailwoman/core/data-root";
+import { isPresent } from "@mailwoman/core/objects";
+import { stableSourceID } from "#adapters/utils";
+import { makeMulberry32, readOATuples } from "#recipes/scaffold";
+import { pick } from "#synthesizers/utils";
+import { alignRow } from "#utils";
+const SOURCE = { zip: dataRootPath("oa-cache", "fr__countrywide.zip"), csv: "fr/countrywide.csv" };
+/**
+ * Ordinal suffixes used in French house numbers (BAN corpus), to cover the "8 bis" sub-mode.
+ */
+const ORDINAL_SUFFIXES = ["bis", "ter", "quater"];
+/**
+ * Probability that a row gets an ordinal suffix injected (matches the golden's ~10-15% rate).
+ */
+const ORDINAL_PROB = 0.12;
+/**
+ * Probability that a locality renders ALL-CAPS (another sub-mode: "SAINTE-LIVRADE-SUR-LOT").
+ */
+const ALLCAPS_PROB = 0.1;
+/**
+ * Stream FR tuples out of the cached OA zip. The countrywide extract is GB-scale, so this reads only as far as `limit`
+ * distinct tuples — the `break` closes the reader and releases the archive. Only keeps rows with a house_number (the
+ * slice's core signal) and a postcode (required for reversed-order rendering to be meaningful; it is also part of this
+ * recipe's dedup key).
+ */
+async function readTuples(limit) {
+    return readOATuples(SOURCE, {
+        limit,
+        requirePostcode: true,
+        dedupIncludesPostcode: true,
+        extra: (fields) => fields,
+    });
+}
+/**
+ * Optionally augment a house_number with a French ordinal suffix ("59 bis", "4 ter"). Appended with a space so it forms
+ * one multi-token house_number string that alignRow can still locate verbatim.
+ */
+function maybeAddOrdinal(random, house_number) {
+    if (random() >= ORDINAL_PROB)
+        return house_number;
+    const suffix = pick(ORDINAL_SUFFIXES, random);
+    // Vary suffix case: "bis" (lower) vs "BIS" (upper) — a real-world split in the golden.
+    return `${house_number} ${random() < 0.5 ? suffix : suffix.toUpperCase()}`;
+}
+/**
+ * Render a tuple in CANONICAL French order: "9 Rue de la Promenade, 01200 Villes".
+ */
+function renderCanonical(hn, street, postcode, locality) {
+    const raw = `${hn} ${street}, ${postcode} ${locality}`;
+    return { raw, components: { house_number: hn, street, postcode, locality } };
+}
+// Reversed layouts, a quarter each: A postcode+city then HN+street; B city, postcode, HN+street; C run-together;
+// D postcode, HN+street, city.
+const REVERSED_VARIANT_A_CUTOFF = 0.25;
+const REVERSED_VARIANT_B_CUTOFF = 0.5;
+const REVERSED_VARIANT_C_CUTOFF = 0.75;
+/**
+ * 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 frOrderRecipe = {
+    name: "fr-order",
+    description: "French reversed-order rows (#560): real OA FR tuples rendered canonical + 4 postcode-first variants",
+    mode: "generate",
+    options: [
+        { flag: "--reversed-fraction <p>", description: "Fraction rendered reversed-order. Default 0.5" },
+        { flag: "--golden", description: "Emit the held-out reversed-order eval slice" },
+    ],
+    async run(opts, write) {
+        if (opts.count == null)
+            throw new Error("fr-order recipe requires --count <N>");
+        const count = opts.count;
+        // Legacy build-fr-order-slice.mjs seeded mulberry32 with the raw seed: `const random = mulberry32(opts.seed)`.
+        // (The omitted diagnostic runSpanCheck used a separate mulberry32(opts.seed + 1) — not part of generation.)
+        const random = makeMulberry32(opts.seed);
+        const source = opts.sourceName ?? "synth-fr-order";
+        const reversedFraction = opts.reversedFraction ?? 0.5;
+        // Over-read from the CSV so the dedup + filter pass can fill `count` rows.
+        const poolLimit = Math.max(count * 8, 40_000);
+        const pool = await readTuples(poolLimit);
+        console.error(`  ${SOURCE.csv}: ${pool.length} unique tuples (capped read)`);
+        if (!pool.length) {
+            throw new Error(`No FR tuples found — is ${SOURCE.zip} present?`);
+        }
+        let emitted = 0;
+        let skipped = 0;
+        let guard = 0;
+        const N = pool.length;
+        while (emitted < count && guard++ < count * 10) {
+            const base = pool[Math.floor(random() * N)];
+            const { street, postcode } = base;
+            const locality = random() < ALLCAPS_PROB ? base.locality.toUpperCase() : base.locality;
+            const house_number = maybeAddOrdinal(random, base.house_number);
+            // Pick canonical vs reversed by --reversed-fraction.
+            const isReversed = random() < reversedFraction;
+            let rendered;
+            if (isReversed) {
+                const variantRoll = random();
+                let raw;
+                if (variantRoll < REVERSED_VARIANT_A_CUTOFF) {
+                    // Variant A: postcode+city as a unit, then HN+street
+                    raw = `${postcode} ${locality}, ${house_number} ${street}`;
+                }
+                else if (variantRoll < REVERSED_VARIANT_B_CUTOFF) {
+                    // Variant B: city, then postcode, then HN+street (comma-separated, postcode isolated)
+                    raw = `${locality}, ${postcode}, ${house_number} ${street}`;
+                }
+                else if (variantRoll < REVERSED_VARIANT_C_CUTOFF) {
+                    // Variant C: no commas — locality HN street postcode (the "run-together" format)
+                    raw = `${locality} ${house_number} ${street} ${postcode}`;
+                }
+                else {
+                    // Variant D: postcode, HN+street, city (reversed top-to-bottom)
+                    raw = `${postcode}, ${house_number} ${street}, ${locality}`;
+                }
+                rendered = { raw, components: { house_number, street, postcode, locality } };
+            }
+            else {
+                rendered = renderCanonical(house_number, street, postcode, locality);
+            }
+            const { raw, components } = rendered;
+            // Safety check: every component must appear verbatim in raw (alignment precondition).
+            const componentValues = Object.values(components).filter(isPresent);
+            if (!componentValues.every((v) => raw.includes(v))) {
+                skipped++;
+                continue;
+            }
+            // --golden: emit per-locale-f1 eval rows ({raw, components, country:"FR"}).
+            if (opts.golden) {
+                write(JSON.stringify({ raw, components, country: "FR" }) + "\n");
+                emitted++;
+                continue;
+            }
+            const sourceID = stableSourceID(source, {
+                street: components.street,
+                house_number: components.house_number,
+                locality: components.locality,
+                postcode: components.postcode,
+            });
+            const canonical = {
+                raw,
+                components,
+                country: "FR",
+                locale: "fr-FR",
+                source,
+                source_id: sourceID,
+                corpus_version: "0.5.0",
+                license: "OpenAddresses FR countrywide tuples, rendered canonical + reversed-order — see ingest SOURCE",
+            };
+            const aligned = alignRow(canonical);
+            if (aligned.kind !== "labeled" || !aligned.row) {
+                skipped++;
+                continue;
+            }
+            write(JSON.stringify({
+                ...aligned.row,
+                synth_method: "fr-order",
+                synth_order: isReversed ? "reversed" : "canonical",
+                synth_base_id: null,
+            }) + "\n");
+            emitted++;
+        }
+        return { emitted, skipped };
+    },
+};
+//# sourceMappingURL=fr-order.js.map

package/out/recipes/fr-order.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fr-order.js","sourceRoot":"","sources":["../../lib/recipes/fr-order.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAA;AACxD,OAAO,EAAE,SAAS,EAAE,MAAM,yBAAyB,CAAA;AAGnD,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAA;AAChD,OAAO,EAAE,cAAc,EAAE,YAAY,EAAqB,MAAM,mBAAmB,CAAA;AACnF,OAAO,EAAE,IAAI,EAAE,MAAM,qBAAqB,CAAA;AAE1C,OAAO,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAA;AAEjC,MAAM,MAAM,GAAG,EAAE,GAAG,EAAE,YAAY,CAAC,UAAU,EAAE,qBAAqB,CAAC,EAAE,GAAG,EAAE,oBAAoB,EAAE,CAAA;AAElG;;GAEG;AACH,MAAM,gBAAgB,GAAsB,CAAC,KAAK,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAA;AACpE;;GAEG;AACH,MAAM,YAAY,GAAG,IAAI,CAAA;AACzB;;GAEG;AACH,MAAM,YAAY,GAAG,GAAG,CAAA;AAYxB;;;;;GAKG;AACH,KAAK,UAAU,UAAU,CAAC,KAAa;IACtC,OAAO,YAAY,CAAC,MAAM,EAAE;QAC3B,KAAK;QACL,eAAe,EAAE,IAAI;QACrB,qBAAqB,EAAE,IAAI;QAC3B,KAAK,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM;KACzB,CAAC,CAAA;AACH,CAAC;AAED;;;GAGG;AACH,SAAS,eAAe,CAAC,MAAoB,EAAE,YAAoB;IAClE,IAAI,MAAM,EAAE,IAAI,YAAY;QAAE,OAAO,YAAY,CAAA;IACjD,MAAM,MAAM,GAAG,IAAI,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAAA;IAE7C,uFAAuF;IACvF,OAAO,GAAG,YAAY,IAAI,MAAM,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,EAAE,EAAE,CAAA;AAC3E,CAAC;AAED;;GAEG;AACH,SAAS,eAAe,CACvB,EAAU,EACV,MAAc,EACd,QAAgB,EAChB,QAAgB;IAEhB,MAAM,GAAG,GAAG,GAAG,EAAE,IAAI,MAAM,KAAK,QAAQ,IAAI,QAAQ,EAAE,CAAA;IAEtD,OAAO,EAAE,GAAG,EAAE,UAAU,EAAE,EAAE,YAAY,EAAE,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,EAAE,CAAA;AAC7E,CAAC;AAED,iHAAiH;AACjH,+BAA+B;AAC/B,MAAM,yBAAyB,GAAG,IAAI,CAAA;AACtC,MAAM,yBAAyB,GAAG,GAAG,CAAA;AACrC,MAAM,yBAAyB,GAAG,IAAI,CAAA;AAEtC;;;GAGG;AACH,MAAM,CAAC,MAAM,aAAa,GAAiB;IAC1C,IAAI,EAAE,UAAU;IAChB,WAAW,EAAE,qGAAqG;IAClH,IAAI,EAAE,UAAU;IAChB,OAAO,EAAE;QACR,EAAE,IAAI,EAAE,yBAAyB,EAAE,WAAW,EAAE,+CAA+C,EAAE;QACjG,EAAE,IAAI,EAAE,UAAU,EAAE,WAAW,EAAE,6CAA6C,EAAE;KAChF;IACD,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,IAAI,IAAI,CAAC,KAAK,IAAI,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAA;QAC/E,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAA;QACxB,+GAA+G;QAC/G,4GAA4G;QAC5G,MAAM,MAAM,GAAG,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACxC,MAAM,MAAM,GAAG,IAAI,CAAC,UAAU,IAAI,gBAAgB,CAAA;QAClD,MAAM,gBAAgB,GAAG,IAAI,CAAC,gBAAgB,IAAI,GAAG,CAAA;QAErD,2EAA2E;QAC3E,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,GAAG,CAAC,EAAE,MAAM,CAAC,CAAA;QAC7C,MAAM,IAAI,GAAG,MAAM,UAAU,CAAC,SAAS,CAAC,CAAA;QAExC,OAAO,CAAC,KAAK,CAAC,KAAK,MAAM,CAAC,GAAG,KAAK,IAAI,CAAC,MAAM,8BAA8B,CAAC,CAAA;QAE5E,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;YAClB,MAAM,IAAI,KAAK,CAAC,2BAA2B,MAAM,CAAC,GAAG,WAAW,CAAC,CAAA;QAClE,CAAC;QAED,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,KAAK,GAAG,CAAC,CAAA;QACb,MAAM,CAAC,GAAG,IAAI,CAAC,MAAM,CAAA;QAErB,OAAO,OAAO,GAAG,KAAK,IAAI,KAAK,EAAE,GAAG,KAAK,GAAG,EAAE,EAAE,CAAC;YAChD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAE,CAAA;YAC5C,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,GAAG,IAAI,CAAA;YACjC,MAAM,QAAQ,GAAG,MAAM,EAAE,GAAG,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAA;YACtF,MAAM,YAAY,GAAG,eAAe,CAAC,MAAM,EAAE,IAAI,CAAC,YAAY,CAAC,CAAA;YAE/D,qDAAqD;YACrD,MAAM,UAAU,GAAG,MAAM,EAAE,GAAG,gBAAgB,CAAA;YAE9C,IAAI,QAA4E,CAAA;YAEhF,IAAI,UAAU,EAAE,CAAC;gBAChB,MAAM,WAAW,GAAG,MAAM,EAAE,CAAA;gBAC5B,IAAI,GAAW,CAAA;gBAEf,IAAI,WAAW,GAAG,yBAAyB,EAAE,CAAC;oBAC7C,qDAAqD;oBACrD,GAAG,GAAG,GAAG,QAAQ,IAAI,QAAQ,KAAK,YAAY,IAAI,MAAM,EAAE,CAAA;gBAC3D,CAAC;qBAAM,IAAI,WAAW,GAAG,yBAAyB,EAAE,CAAC;oBACpD,sFAAsF;oBACtF,GAAG,GAAG,GAAG,QAAQ,KAAK,QAAQ,KAAK,YAAY,IAAI,MAAM,EAAE,CAAA;gBAC5D,CAAC;qBAAM,IAAI,WAAW,GAAG,yBAAyB,EAAE,CAAC;oBACpD,iFAAiF;oBACjF,GAAG,GAAG,GAAG,QAAQ,IAAI,YAAY,IAAI,MAAM,IAAI,QAAQ,EAAE,CAAA;gBAC1D,CAAC;qBAAM,CAAC;oBACP,gEAAgE;oBAChE,GAAG,GAAG,GAAG,QAAQ,KAAK,YAAY,IAAI,MAAM,KAAK,QAAQ,EAAE,CAAA;gBAC5D,CAAC;gBAED,QAAQ,GAAG,EAAE,GAAG,EAAE,UAAU,EAAE,EAAE,YAAY,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,EAAE,CAAA;YAC7E,CAAC;iBAAM,CAAC;gBACP,QAAQ,GAAG,eAAe,CAAC,YAAY,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAA;YACrE,CAAC;YAED,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,GAAG,QAAQ,CAAA;YAEpC,sFAAsF;YACtF,MAAM,eAAe,GAAG,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAA;YAEnE,IAAI,CAAC,eAAe,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;gBACpD,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,4EAA4E;YAC5E,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;gBACjB,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,UAAU,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,GAAG,IAAI,CAAC,CAAA;gBAEhE,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,MAAM,QAAQ,GAAG,cAAc,CAAC,MAAM,EAAE;gBACvC,MAAM,EAAE,UAAU,CAAC,MAAM;gBACzB,YAAY,EAAE,UAAU,CAAC,YAAY;gBACrC,QAAQ,EAAE,UAAU,CAAC,QAAQ;gBAC7B,QAAQ,EAAE,UAAU,CAAC,QAAQ;aAC7B,CAAC,CAAA;YAEF,MAAM,SAAS,GAAiB;gBAC/B,GAAG;gBACH,UAAU;gBACV,OAAO,EAAE,IAAI;gBACb,MAAM,EAAE,OAAO;gBACf,MAAM;gBACN,SAAS,EAAE,QAAQ;gBACnB,cAAc,EAAE,OAAO;gBACvB,OAAO,EAAE,8FAA8F;aACvG,CAAA;YAED,MAAM,OAAO,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAA;YAEnC,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,UAAU;gBACxB,WAAW,EAAE,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,WAAW;gBAClD,aAAa,EAAE,IAAI;aACnB,CAAC,GAAG,IAAI,CACT,CAAA;YAED,OAAO,EAAE,CAAA;QACV,CAAC;QAED,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAC5B,CAAC;CACD,CAAA"}

package/out/recipes/german.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=german.d.ts.map

package/out/recipes/german.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"german.d.ts","sourceRoot":"","sources":["../../lib/recipes/german.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AASH,OAAO,EAAwC,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAA;AA0D3F;;;;;;;;;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/recipes/german.js

@@ -0,0 +1,288 @@
+/**
+ * @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 { DatabaseClient } from "@mailwoman/sqlite/client";
+import { stableSourceID } from "#adapters/utils";
+import { makeMulberry32, readZippedCSVRecords } from "#recipes/scaffold";
+import { synthesizeGermanRow } from "#synthesizers/german";
+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=german.js.map

package/out/recipes/german.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"german.js","sourceRoot":"","sources":["../../lib/recipes/german.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAA;AACxD,OAAO,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAA;AAEvD,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAA;AAGzD,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAA;AAChD,OAAO,EAAE,cAAc,EAAE,oBAAoB,EAAqB,MAAM,mBAAmB,CAAA;AAC3F,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/recipes/house-venue.d.ts

@@ -0,0 +1,18 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `house-venue` slice recipe — synthetic house_number+venue+street co-occurrence rows: tuples →
+ *   {@link synthesizeHouseVenueRow} → aligned LabeledRow. The v0.6.3 corrective companion to the
+ *   no-street slice: every row carries BOTH house_number AND venue, restoring the house_number
+ *   signal that no-street's distributional shift cost the model. Ported from
+ *   scripts/build-house-venue-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 houseVenueRecipe: CorpusRecipe;
+//# sourceMappingURL=house-venue.d.ts.map

package/out/recipes/house-venue.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"house-venue.d.ts","sourceRoot":"","sources":["../../lib/recipes/house-venue.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAMN,KAAK,YAAY,EACjB,MAAM,mBAAmB,CAAA;AAG1B;;;GAGG;AACH,eAAO,MAAM,gBAAgB,EAAE,YAiE9B,CAAA"}