@mailwoman/corpus 4.16.2 → 5.1.0

package/out/src/shard-recipes/german.js

@@ -0,0 +1,170 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `german` shard 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-shard.mjs.
+ *
+ *   ORDER ROBUSTNESS (2026-06-06): mixing the two renderings stops a native-only shard from teaching
+ *   German order so well it reads the US/feed-order eval as a "collapse". See
+ *   docs/articles/evals/2026-06-06-anchor-pilot.md (the order-artifact correction).
+ */
+import { spawnSync } from "node:child_process";
+import { stableSourceID } from "../adapter.js";
+import { alignRow } from "../align.js";
+import { synthesizeGermanRow } from "../synthesize-german.js";
+import { makeMulberry32 } from "./scaffold.js";
+// `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: "/tmp/oa-cache/de__berlin.zip", csv: "de/berlin.csv", region: "Berlin" },
+    { zip: "/tmp/oa-cache/de__sn__statewide.zip", csv: "de/sn/statewide.csv", region: "Sachsen" },
+];
+/** Minimal RFC-4180-ish splitter (handles quoted fields). */
+function splitCSV(line) {
+    const out = [];
+    let cur = "";
+    let inQ = false;
+    for (let i = 0; i < line.length; i++) {
+        const c = line[i];
+        if (inQ) {
+            if (c === '"') {
+                if (line[i + 1] === '"') {
+                    cur += '"';
+                    i++;
+                }
+                else
+                    inQ = false;
+            }
+            else
+                cur += c;
+        }
+        else if (c === '"')
+            inQ = true;
+        else if (c === ",") {
+            out.push(cur);
+            cur = "";
+        }
+        else
+            cur += c;
+    }
+    out.push(cur);
+    return out;
+}
+/** Stream real German tuples out of a cached OA zip (buffered `unzip -p`). */
+function readGermanTuples(source) {
+    const r = spawnSync("unzip", ["-p", source.zip, source.csv], { maxBuffer: 1024 * 1024 * 1024, encoding: "buffer" });
+    if (r.status !== 0) {
+        console.error(`  WARN: unzip failed for ${source.zip} (status ${r.status})`);
+        return [];
+    }
+    const lines = r.stdout.toString("utf8").split(/\r?\n/);
+    if (lines.length < 2)
+        return [];
+    const header = splitCSV(lines[0]).map((h) => h.trim().toLowerCase());
+    const idx = (name) => header.indexOf(name);
+    const iNum = idx("number"), iStreet = idx("street"), iCity = idx("city"), iRegion = idx("region"), iPost = idx("postcode");
+    const get = (cells, i) => (i >= 0 && i < cells.length ? (cells[i] ?? "").trim() : "");
+    const tuples = [];
+    const seen = new Set();
+    for (let li = 1; li < lines.length; li++) {
+        const lineStr = lines[li];
+        if (!lineStr)
+            continue;
+        const cells = splitCSV(lineStr);
+        const street = get(cells, iStreet);
+        const locality = get(cells, iCity);
+        if (!street || !locality)
+            continue;
+        const house_number = get(cells, iNum);
+        const postcode = get(cells, iPost);
+        // OA's REGION column is empty for DE — fall back to the source's Bundesland (set per file).
+        const region = get(cells, iRegion) || source.region || "";
+        const key = `${house_number}|${street}|${locality}|${postcode}`.toLowerCase();
+        if (seen.has(key))
+            continue;
+        seen.add(key);
+        tuples.push({ house_number, street, locality, region, postcode });
+    }
+    return tuples;
+}
+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" }],
+    async run(opts, write) {
+        // Emit PRNG: the legacy build-german-shard.mjs seeded mulberry32(opts.seed).
+        const random = makeMulberry32(opts.seed);
+        const source = opts.sourceName ?? "synth-german";
+        const intlFraction = opts.intlFraction ?? 0.4;
+        if (!(intlFraction >= 0 && intlFraction <= 1)) {
+            throw new Error(`--intl-fraction must be in [0, 1], got ${intlFraction}`);
+        }
+        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 = 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 === 0) {
+            throw new Error("No German tuples found — are the cached zips present in /tmp/oa-cache?");
+        }
+        let emitted = 0;
+        let skipped = 0;
+        let guard = 0;
+        const N = pool.length;
+        while (emitted < count && guard++ < count * 6) {
+            const base = 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";
+            const synth = synthesizeGermanRow(base, { random, order });
+            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_base_id: null }) + "\n");
+            emitted++;
+        }
+        return { emitted, skipped };
+    },
+};
+//# sourceMappingURL=german.js.map

package/out/src/shard-recipes/german.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"german.js","sourceRoot":"","sources":["../../../src/shard-recipes/german.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAA;AAE9C,OAAO,EAAE,cAAc,EAAE,MAAM,eAAe,CAAA;AAC9C,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAA;AACtC,OAAO,EAAE,mBAAmB,EAAwB,MAAM,yBAAyB,CAAA;AACnF,OAAO,EAAE,cAAc,EAAoB,MAAM,eAAe,CAAA;AAWhE,sGAAsG;AACtG,wGAAwG;AACxG,iGAAiG;AACjG,MAAM,OAAO,GAAmB;IAC/B,EAAE,GAAG,EAAE,8BAA8B,EAAE,GAAG,EAAE,eAAe,EAAE,MAAM,EAAE,QAAQ,EAAE;IAC/E,EAAE,GAAG,EAAE,qCAAqC,EAAE,GAAG,EAAE,qBAAqB,EAAE,MAAM,EAAE,SAAS,EAAE;CAC7F,CAAA;AAED,6DAA6D;AAC7D,SAAS,QAAQ,CAAC,IAAY;IAC7B,MAAM,GAAG,GAAa,EAAE,CAAA;IACxB,IAAI,GAAG,GAAG,EAAE,CAAA;IACZ,IAAI,GAAG,GAAG,KAAK,CAAA;IAEf,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAE,CAAA;QAElB,IAAI,GAAG,EAAE,CAAC;YACT,IAAI,CAAC,KAAK,GAAG,EAAE,CAAC;gBACf,IAAI,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,GAAG,EAAE,CAAC;oBACzB,GAAG,IAAI,GAAG,CAAA;oBACV,CAAC,EAAE,CAAA;gBACJ,CAAC;;oBAAM,GAAG,GAAG,KAAK,CAAA;YACnB,CAAC;;gBAAM,GAAG,IAAI,CAAC,CAAA;QAChB,CAAC;aAAM,IAAI,CAAC,KAAK,GAAG;YAAE,GAAG,GAAG,IAAI,CAAA;aAC3B,IAAI,CAAC,KAAK,GAAG,EAAE,CAAC;YACpB,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;YACb,GAAG,GAAG,EAAE,CAAA;QACT,CAAC;;YAAM,GAAG,IAAI,CAAC,CAAA;IAChB,CAAC;IACD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;IAEb,OAAO,GAAG,CAAA;AACX,CAAC;AAED,8EAA8E;AAC9E,SAAS,gBAAgB,CAAC,MAAoB;IAC7C,MAAM,CAAC,GAAG,SAAS,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,MAAM,CAAC,GAAG,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,GAAG,IAAI,GAAG,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC,CAAA;IAEnH,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpB,OAAO,CAAC,KAAK,CAAC,4BAA4B,MAAM,CAAC,GAAG,YAAY,CAAC,CAAC,MAAM,GAAG,CAAC,CAAA;QAE5E,OAAO,EAAE,CAAA;IACV,CAAC;IACD,MAAM,KAAK,GAAG,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAA;IAEtD,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,EAAE,CAAA;IAC/B,MAAM,MAAM,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAA;IACrE,MAAM,GAAG,GAAG,CAAC,IAAY,EAAU,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAA;IAC1D,MAAM,IAAI,GAAG,GAAG,CAAC,QAAQ,CAAC,EACzB,OAAO,GAAG,GAAG,CAAC,QAAQ,CAAC,EACvB,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,EACnB,OAAO,GAAG,GAAG,CAAC,QAAQ,CAAC,EACvB,KAAK,GAAG,GAAG,CAAC,UAAU,CAAC,CAAA;IACxB,MAAM,GAAG,GAAG,CAAC,KAAe,EAAE,CAAS,EAAU,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAA;IAC/G,MAAM,MAAM,GAAsB,EAAE,CAAA;IACpC,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAE9B,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,MAAM,EAAE,EAAE,EAAE,EAAE,CAAC;QAC1C,MAAM,OAAO,GAAG,KAAK,CAAC,EAAE,CAAC,CAAA;QAEzB,IAAI,CAAC,OAAO;YAAE,SAAQ;QACtB,MAAM,KAAK,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAA;QAC/B,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;QAClC,MAAM,QAAQ,GAAG,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,CAAA;QAElC,IAAI,CAAC,MAAM,IAAI,CAAC,QAAQ;YAAE,SAAQ;QAClC,MAAM,YAAY,GAAG,GAAG,CAAC,KAAK,EAAE,IAAI,CAAC,CAAA;QACrC,MAAM,QAAQ,GAAG,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,CAAA;QAClC,4FAA4F;QAC5F,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,MAAM,IAAI,EAAE,CAAA;QACzD,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,MAAM,CAAC,MAAM,YAAY,GAAgB;IACxC,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,6FAA6F;IAC1G,IAAI,EAAE,UAAU;IAChB,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,WAAW,EAAE,oDAAoD,EAAE,CAAC;IAC7G,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;QAE7C,IAAI,CAAC,CAAC,YAAY,IAAI,CAAC,IAAI,YAAY,IAAI,CAAC,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,KAAK,CAAC,0CAA0C,YAAY,EAAE,CAAC,CAAA;QAC1E,CAAC;QACD,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,gBAAgB,CAAC,CAAC,CAAC,CAAA;YAC7B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,MAAM,gBAAgB,CAAC,CAAA;YAEtD,KAAK,MAAM,CAAC,IAAI,CAAC;gBAAE,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA,CAAC,iEAAiE;QAClG,CAAC;QAED,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACvB,MAAM,IAAI,KAAK,CAAC,wEAAwE,CAAC,CAAA;QAC1F,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,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAE,CAAA;YAC5C,iGAAiG;YACjG,2EAA2E;YAC3E,MAAM,KAAK,GAAG,MAAM,EAAE,GAAG,YAAY,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,QAAQ,CAAA;YAClE,MAAM,KAAK,GAAG,mBAAmB,CAAC,IAAI,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,CAAA;YAE1D,IAAI,CAAC,KAAK,EAAE,CAAC;gBACZ,OAAO,EAAE,CAAA;gBACT,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;gBACpG,OAAO,EAAE,CAAA;gBACT,SAAQ;YACT,CAAC;YACD,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;YACF,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;YACD,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;gBACT,SAAQ;YACT,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,YAAY,EAAE,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,GAAG,IAAI,CAAC,CAAA;YACjH,OAAO,EAAE,CAAA;QACV,CAAC;QAED,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAC5B,CAAC;CACD,CAAA"}

package/out/src/shard-recipes/house-venue.d.ts

@@ -0,0 +1,14 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `house-venue` shard recipe — synthetic house_number+venue+street co-occurrence rows: tuples →
+ *   {@link synthesizeHouseVenueRow} → aligned LabeledRow. The v0.6.3 corrective companion to the
+ *   no-street shard: 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-shard.mjs.
+ */
+import { type ShardRecipe } from "./scaffold.js";
+export declare const houseVenueRecipe: ShardRecipe;
+//# sourceMappingURL=house-venue.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"house-venue.d.ts","sourceRoot":"","sources":["../../../src/shard-recipes/house-venue.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAGH,OAAO,EAAqD,KAAK,WAAW,EAAE,MAAM,eAAe,CAAA;AAInG,eAAO,MAAM,gBAAgB,EAAE,WAyD9B,CAAA"}

package/out/src/shard-recipes/house-venue.js

@@ -0,0 +1,66 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `house-venue` shard recipe — synthetic house_number+venue+street co-occurrence rows: tuples →
+ *   {@link synthesizeHouseVenueRow} → aligned LabeledRow. The v0.6.3 corrective companion to the
+ *   no-street shard: 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-shard.mjs.
+ */
+import { synthesizeHouseVenueRow } from "../synthesize-house-venue.js";
+import { alignAndWrite, makeLcg, readTuples, shardSourceID } from "./scaffold.js";
+const LICENSE = "Synthetic — derived from CC-BY / public-domain input tuples";
+export const houseVenueRecipe = {
+    name: "house-venue",
+    description: "House_number+venue co-occurrence rows: tuples → synthesizeHouseVenueRow → aligned LabeledRow",
+    mode: "tuples",
+    async run(opts, write) {
+        if (!opts.input)
+            throw new Error("house-venue recipe requires --input <tuples.jsonl>");
+        // Legacy build-house-venue-shard.mjs seeded the LCG via makeRandom(opts.seed) (s = seed).
+        const random = makeLcg(opts.seed);
+        const source = opts.sourceName ?? "synth-house-venue";
+        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 = synthesizeHouseVenueRow(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: shardSourceID(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=house-venue.js.map

package/out/src/shard-recipes/house-venue.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"house-venue.js","sourceRoot":"","sources":["../../../src/shard-recipes/house-venue.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,uBAAuB,EAA4B,MAAM,8BAA8B,CAAA;AAChG,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,UAAU,EAAE,aAAa,EAAoB,MAAM,eAAe,CAAA;AAEnG,MAAM,OAAO,GAAG,6DAA6D,CAAA;AAE7E,MAAM,CAAC,MAAM,gBAAgB,GAAgB;IAC5C,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE,8FAA8F;IAC3G,IAAI,EAAE,QAAQ;IACd,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,IAAI,CAAC,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAA;QACtF,0FAA0F;QAC1F,MAAM,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACjC,MAAM,MAAM,GAAG,IAAI,CAAC,UAAU,IAAI,mBAAmB,CAAA;QACrD,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;gBACT,SAAQ;YACT,CAAC;YAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;gBACxC,MAAM,KAAK,GAAG,uBAAuB,CAAC,KAA4B,EAAE,EAAE,MAAM,EAAE,CAAC,CAAA;gBAE/E,IAAI,CAAC,KAAK,EAAE,CAAC;oBACZ,OAAO,EAAE,CAAA;oBACT,SAAQ;gBACT,CAAC;gBACD,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;oBAAE,OAAO,EAAE,CAAA;;oBACZ,OAAO,EAAE,CAAA;YACf,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}

package/out/src/shard-recipes/index.d.ts

@@ -0,0 +1,18 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   The synthetic-corpus SHARD RECIPE registry. Each recipe ({@link ShardRecipe}) is what one legacy
+ *   `build-*-shard.mjs` script used to be; the `mailwoman corpus shard <recipe>` command dispatches
+ *   to them. A new shard is a new entry here, not a new script.
+ */
+import type { ShardRecipe } from "./scaffold.js";
+export * from "./scaffold.js";
+/** Recipe name → recipe. */
+export declare const SHARD_RECIPES: ReadonlyMap<string, ShardRecipe>;
+/** Look up a recipe by its `<recipe>` name. */
+export declare function getShardRecipe(name: string): ShardRecipe | undefined;
+/** All recipes (for `--list` / help). */
+export declare function listShardRecipes(): readonly ShardRecipe[];
+//# sourceMappingURL=index.d.ts.map

package/out/src/shard-recipes/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/shard-recipes/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAeH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,eAAe,CAAA;AAMhD,cAAc,eAAe,CAAA;AAuB7B,4BAA4B;AAC5B,eAAO,MAAM,aAAa,EAAE,WAAW,CAAC,MAAM,EAAE,WAAW,CAA4C,CAAA;AAEvG,+CAA+C;AAC/C,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,SAAS,CAEpE;AAED,yCAAyC;AACzC,wBAAgB,gBAAgB,IAAI,SAAS,WAAW,EAAE,CAEzD"}

package/out/src/shard-recipes/index.js

@@ -0,0 +1,58 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   The synthetic-corpus SHARD RECIPE registry. Each recipe ({@link ShardRecipe}) is what one legacy
+ *   `build-*-shard.mjs` script used to be; the `mailwoman corpus shard <recipe>` command dispatches
+ *   to them. A new shard is a new entry here, not a new script.
+ */
+import { anchorAbsorptionRecipe } from "./anchor-absorption.js";
+import { boundaryStressRecipe } from "./boundary-stress.js";
+import { countryBalancedRecipe } from "./country-balanced.js";
+import { frAdminSplitRecipe } from "./fr-admin-split.js";
+import { frBareStreetRecipe } from "./fr-bare-street.js";
+import { frOrderRecipe } from "./fr-order.js";
+import { germanRecipe } from "./german.js";
+import { houseVenueRecipe } from "./house-venue.js";
+import { intersectionRecipe } from "./intersection.js";
+import { localeRecipe } from "./locale.js";
+import { noStreetRecipe } from "./no-street.js";
+import { poBoxCedexRecipe } from "./po-box-cedex.js";
+import { poBoxRecipe } from "./po-box.js";
+import { streetAffixRecipe } from "./street-affix.js";
+import { streetBareRecipe } from "./street-bare.js";
+import { streetRecipe } from "./street.js";
+import { unitRecipe } from "./unit.js";
+export * from "./scaffold.js";
+/** Every registered recipe, in display order. */
+const RECIPES = [
+    streetRecipe,
+    streetBareRecipe,
+    streetAffixRecipe,
+    noStreetRecipe,
+    houseVenueRecipe,
+    poBoxRecipe,
+    poBoxCedexRecipe,
+    unitRecipe,
+    intersectionRecipe,
+    germanRecipe,
+    localeRecipe,
+    frOrderRecipe,
+    frAdminSplitRecipe,
+    frBareStreetRecipe,
+    countryBalancedRecipe,
+    boundaryStressRecipe,
+    anchorAbsorptionRecipe,
+];
+/** Recipe name → recipe. */
+export const SHARD_RECIPES = new Map(RECIPES.map((r) => [r.name, r]));
+/** Look up a recipe by its `<recipe>` name. */
+export function getShardRecipe(name) {
+    return SHARD_RECIPES.get(name);
+}
+/** All recipes (for `--list` / help). */
+export function listShardRecipes() {
+    return RECIPES;
+}
+//# sourceMappingURL=index.js.map

package/out/src/shard-recipes/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/shard-recipes/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAA;AAC/D,OAAO,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAA;AAC3D,OAAO,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAA;AAC7D,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAA;AACxD,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAA;AACxD,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAA;AAC7C,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAA;AAC1C,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAA;AACnD,OAAO,EAAE,kBAAkB,EAAE,MAAM,mBAAmB,CAAA;AACtD,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAA;AAC1C,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAA;AAC/C,OAAO,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAA;AACpD,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAA;AAEzC,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAA;AACrD,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAA;AACnD,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAA;AAC1C,OAAO,EAAE,UAAU,EAAE,MAAM,WAAW,CAAA;AAEtC,cAAc,eAAe,CAAA;AAE7B,iDAAiD;AACjD,MAAM,OAAO,GAA2B;IACvC,YAAY;IACZ,gBAAgB;IAChB,iBAAiB;IACjB,cAAc;IACd,gBAAgB;IAChB,WAAW;IACX,gBAAgB;IAChB,UAAU;IACV,kBAAkB;IAClB,YAAY;IACZ,YAAY;IACZ,aAAa;IACb,kBAAkB;IAClB,kBAAkB;IAClB,qBAAqB;IACrB,oBAAoB;IACpB,sBAAsB;CACtB,CAAA;AAED,4BAA4B;AAC5B,MAAM,CAAC,MAAM,aAAa,GAAqC,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAA;AAEvG,+CAA+C;AAC/C,MAAM,UAAU,cAAc,CAAC,IAAY;IAC1C,OAAO,aAAa,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;AAC/B,CAAC;AAED,yCAAyC;AACzC,MAAM,UAAU,gBAAgB;IAC/B,OAAO,OAAO,CAAA;AACf,CAAC"}

package/out/src/shard-recipes/intersection.d.ts

@@ -0,0 +1,38 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `intersection` shard recipe — the REAL-pair intersection training shard (#487). The model scored
+ *   0.0 on intersection_a/b because the training mix had ZERO intersection-labeled rows. This is
+ *   the missing data. Ported from scripts/build-intersection-shard.mjs.
+ *
+ *   STREET PAIRS ARE REAL: the same TIGER 2023 EDGES extraction as the eval builder
+ *   (scripts/eval/build-intersection-real.ts) — a node where two road edges (MTFCC S1*) with
+ *   distinct FULLNAMEs meet is a real crossing. Real pairs avoid teaching fake street-street
+ *   co-occurrences.
+ *
+ *   LEAKAGE POLICY (mirrors the affix shard's VT discipline):
+ *
+ *   - TRAIN counties: Cook IL (grid city) + Morris NJ (suburb).
+ *   - GOLDEN (`--golden`) county: Washington VT (rural) ONLY — the corpus defaultHoldout state.
+ *   - Every crossing in data/eval/external/intersection-real.jsonl is excluded from BOTH modes, by node
+ *       id AND by order-insensitive name pair (the eval shares all three counties).
+ *
+ *   RENDERING: junction-format variety — padded/TIGHT `&` and `/`, `and`, `at`, `@`, leading-phrase
+ *   `corner of` / `intersection of` — crossed with tails (bare / `, ST` / `, ST ZIP` / `, City, ST
+ *   [ZIP]`) and case variants. ZIPs are the crossing's own TIGER edge ZIPL (real); the locality
+ *   tail comes from the OA Cook-county ZIP→city majority map.
+ *
+ *   AUDIT: every emitted row is label-checked on the RAW SURFACE via the #519 char-offset span
+ *   triple. Any violation FAILS the build (throws). A JSON audit report lands next to the output.
+ *
+ *   External inputs (`--edges-dir`, opts.edgesDir; both already on disk — do not re-download):
+ *
+ *   - <edges-dir>/tl_2023_{17031,34027,50023}_edges.shp (unzipped TIGER 2023 EDGES; default
+ *       /tmp/tiger-edges)
+ *   - /tmp/oa-cache/us__il__cook.zip (ZIP→city tails)
+ */
+import { type ShardRecipe } from "./scaffold.js";
+export declare const intersectionRecipe: ShardRecipe;
+//# sourceMappingURL=intersection.d.ts.map

package/out/src/shard-recipes/intersection.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"intersection.d.ts","sourceRoot":"","sources":["../../../src/shard-recipes/intersection.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAkCG;AAYH,OAAO,EAAkB,KAAK,WAAW,EAAE,MAAM,eAAe,CAAA;AAsXhE,eAAO,MAAM,kBAAkB,EAAE,WAkKhC,CAAA"}