@mailwoman/corpus 9.3.0 → 10.0.0

package/out/recipes/street/bare.js

@@ -0,0 +1,87 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `street-bare` slice recipe — BARE-street rows (the v0.8.0 harness change, 2026-06-05). The
+ *   `functional.test.ts` cluster (bare street names — "10th Ave", "Main St", "1 Main Pl") was
+ *   mislabeled `locality` because {@link synthesizeStreetRow} only ever emitted streets WITH a ",
+ *   City, ST ZIP" tail. This recipe emits streets BARE (`--bare-prob`, default 0.6) — no tail, only
+ *   `street_prefix`/`street`/`street_suffix` (+ optional `house_number` at `--hn-prob`, default
+ *   0.85) — over the built-in {@link DEFAULT_US_BASES} pool (round-robin). Generate-mode, US-only,
+ *   in-distribution (no German-collapse risk). Ported from scripts/build-street-bare-slice.mjs.
+ *
+ *   Byte-fidelity: the legacy script seeded its own mulberry32 from `--seed`
+ *   (`mulberry32(opts.seed)`); this recipe re-creates the SAME generator
+ *   (`makeMulberry32(opts.seed)`) and preserves the synthesis call order exactly, so `--seed N`
+ *   reproduces the legacy run byte-for-byte.
+ */
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
+import { sliceSourceID } from "#recipes/scaffold";
+import { DEFAULT_US_BASES } from "#synthesizers/intersection";
+import { synthesizeStreetRow } from "#synthesizers/street";
+import { alignRow } from "#utils";
+/**
+ * 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 streetBareRecipe = {
+    name: "street-bare",
+    description: "Bare-street rows (US): DEFAULT_US_BASES → synthesizeStreetRow (bare) → aligned LabeledRow",
+    mode: "generate",
+    options: [
+        { flag: "--bare-prob <p>", description: "P(emit the street BARE — no city/region/postcode tail). Default 0.6" },
+        { flag: "--hn-prob <p>", description: "P(emit a house number). Default 0.85" },
+    ],
+    async run(opts, write) {
+        const random = makeMulberry32(opts.seed);
+        const count = opts.count ?? 3000;
+        const bareProb = opts.bareProb ?? 0.6;
+        const hnProb = opts.hnProb ?? 0.85;
+        const source = opts.sourceName ?? "synth-street-bare";
+        let emitted = 0;
+        let skipped = 0;
+        let guard = 0;
+        while (emitted < count && guard++ < count * 5) {
+            const base = DEFAULT_US_BASES[emitted % DEFAULT_US_BASES.length];
+            const synth = synthesizeStreetRow(base, {
+                random,
+                bareProb,
+                includeHouseNumberProb: hnProb,
+            });
+            if (!synth) {
+                skipped++;
+                continue;
+            }
+            const isBare = synth.components.region === undefined;
+            const canonical = {
+                raw: synth.raw,
+                components: synth.components,
+                country: base.country,
+                locale: synth.locale,
+                source,
+                source_id: sliceSourceID(source, {
+                    street: synth.components.street,
+                    street_suffix: synth.components.street_suffix,
+                    house_number: synth.components.house_number,
+                    bare: String(isBare),
+                    n: String(emitted),
+                }),
+                corpus_version: "0.4.0",
+                license: "Synthetic — US street templates, public-domain street/city pools",
+            };
+            // Strict labeled-only check (matches the legacy builder): alignRow always returns a `row`
+            // (labeled OR quarantined), so the scaffold's `alignAndWrite` would write quarantined rows
+            // too — call alignRow directly and skip anything not "labeled".
+            const aligned = alignRow(canonical);
+            if (aligned.kind !== "labeled" || !aligned.row) {
+                skipped++;
+                continue;
+            }
+            write(JSON.stringify({ ...aligned.row, synth_method: "street-bare", synth_base_id: null }) + "\n");
+            emitted++;
+        }
+        return { emitted, skipped };
+    },
+};
+//# sourceMappingURL=bare.js.map

package/out/recipes/street/bare.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"bare.js","sourceRoot":"","sources":["../../../lib/recipes/street/bare.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AAEpE,OAAO,EAAE,aAAa,EAA6C,MAAM,mBAAmB,CAAA;AAC5F,OAAO,EAAE,gBAAgB,EAAE,MAAM,4BAA4B,CAAA;AAC7D,OAAO,EAAE,mBAAmB,EAAwB,MAAM,sBAAsB,CAAA;AAChF,OAAO,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAA;AAEjC;;;GAGG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAiB;IAC7C,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE,2FAA2F;IACxG,IAAI,EAAE,UAAU;IAChB,OAAO,EAAE;QACR,EAAE,IAAI,EAAE,iBAAiB,EAAE,WAAW,EAAE,qEAAqE,EAAE;QAC/G,EAAE,IAAI,EAAE,eAAe,EAAE,WAAW,EAAE,sCAAsC,EAAE;KAC9E;IACD,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,MAAM,MAAM,GAAG,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACxC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,IAAI,CAAA;QAChC,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,GAAG,CAAA;QACrC,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,IAAI,IAAI,CAAA;QAClC,MAAM,MAAM,GAAG,IAAI,CAAC,UAAU,IAAI,mBAAmB,CAAA;QAErD,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,KAAK,GAAG,CAAC,CAAA;QAEb,OAAO,OAAO,GAAG,KAAK,IAAI,KAAK,EAAE,GAAG,KAAK,GAAG,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,GAAG,gBAAgB,CAAC,MAAM,CAAE,CAAA;YAEjE,MAAM,KAAK,GAAG,mBAAmB,CAAC,IAAuB,EAAE;gBAC1D,MAAM;gBACN,QAAQ;gBACR,sBAAsB,EAAE,MAAM;aAC9B,CAAC,CAAA;YAEF,IAAI,CAAC,KAAK,EAAE,CAAC;gBACZ,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,MAAM,MAAM,GAAG,KAAK,CAAC,UAAU,CAAC,MAAM,KAAK,SAAS,CAAA;YAEpD,MAAM,SAAS,GAAsB;gBACpC,GAAG,EAAE,KAAK,CAAC,GAAG;gBACd,UAAU,EAAE,KAAK,CAAC,UAAU;gBAC5B,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,MAAM;gBACN,SAAS,EAAE,aAAa,CAAC,MAAM,EAAE;oBAChC,MAAM,EAAE,KAAK,CAAC,UAAU,CAAC,MAAM;oBAC/B,aAAa,EAAE,KAAK,CAAC,UAAU,CAAC,aAAa;oBAC7C,YAAY,EAAE,KAAK,CAAC,UAAU,CAAC,YAAY;oBAC3C,IAAI,EAAE,MAAM,CAAC,MAAM,CAAC;oBACpB,CAAC,EAAE,MAAM,CAAC,OAAO,CAAC;iBAClB,CAAC;gBACF,cAAc,EAAE,OAAO;gBACvB,OAAO,EAAE,kEAAkE;aAC3E,CAAA;YAED,0FAA0F;YAC1F,2FAA2F;YAC3F,gEAAgE;YAChE,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,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,YAAY,EAAE,aAAa,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,GAAG,IAAI,CAAC,CAAA;YAElG,OAAO,EAAE,CAAA;QACV,CAAC;QAED,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAC5B,CAAC;CACD,CAAA"}

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

@@ -1 +1 @@
-{"version":3,"file":"street.d.ts","sourceRoot":"","sources":["../../lib/recipes/street.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAGH,OAAO,EAAsC,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAA;AAGzF;;;GAGG;AACH,eAAO,MAAM,YAAY,EAAE,YA+D1B,CAAA"}
+{"version":3,"file":"street.d.ts","sourceRoot":"","sources":["../../lib/recipes/street.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAKH,OAAO,EAA6B,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAA;AAGhF;;;GAGG;AACH,eAAO,MAAM,YAAY,EAAE,YA+D1B,CAAA"}

package/out/recipes/street.js

@@ -6,8 +6,9 @@
  *   `street` slice recipe — synthetic street-decomposition rows for Stage 3 (US-only): tuples →
  *   {@link synthesizeStreetRow} → aligned LabeledRow. Ported from scripts/build-street-slice.mjs.
  */
+import { makeLcg } from "@mailwoman/core/utils";
 import { stableSourceID } from "#adapters/utils";
-import { alignAndWrite, makeLcg, readTuples } from "#recipes/scaffold";
+import { alignAndWrite, readTuples } from "#recipes/scaffold";
 import { synthesizeStreetRow } from "#synthesizers/street";
 /**
  * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,

package/out/recipes/street.js.map

@@ -1 +1 @@
-{"version":3,"file":"street.js","sourceRoot":"","sources":["../../lib/recipes/street.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAA;AAChD,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,UAAU,EAAqB,MAAM,mBAAmB,CAAA;AACzF,OAAO,EAAE,mBAAmB,EAAwB,MAAM,sBAAsB,CAAA;AAEhF;;;GAGG;AACH,MAAM,CAAC,MAAM,YAAY,GAAiB;IACzC,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,mFAAmF;IAChG,IAAI,EAAE,QAAQ;IACd,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,yBAAyB,EAAE,WAAW,EAAE,sCAAsC,EAAE,CAAC;IACnG,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,IAAI,CAAC,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,+CAA+C,CAAC,CAAA;QACjF,MAAM,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACjC,MAAM,sBAAsB,GAAG,IAAI,CAAC,eAAe,IAAI,IAAI,CAAA;QAC3D,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QAEf,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAClD,IAAI,EAAE,CAAA;YAEN,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBAC3E,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;gBAC5B,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;gBACxC,MAAM,KAAK,GAAG,mBAAmB,CAAC,KAAwB,EAAE,EAAE,MAAM,EAAE,sBAAsB,EAAE,CAAC,CAAA;gBAE/F,IAAI,CAAC,KAAK;oBAAE,SAAQ;gBAEpB,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,EAAE,cAAc;oBACtB,SAAS,EAAE,cAAc,CAAC,cAAc,EAAE;wBACzC,QAAQ,EAAE,GAAG,KAAK,CAAC,QAAQ,IAAI,CAAC,EAAE;wBAClC,MAAM,EAAE,KAAK,CAAC,MAAM;wBACpB,QAAQ,EAAE,KAAK,CAAC,QAAQ;wBACxB,OAAO,EAAE,KAAK,CAAC,OAAO;qBACtB,CAAC;oBACF,cAAc,EAAE,OAAO;oBACvB,OAAO,EAAE,2DAA2D;iBACpE,EACD,eAAe,CACf,CAAA;gBAED,IAAI,EAAE,EAAE,CAAC;oBACR,OAAO,EAAE,CAAA;gBACV,CAAC;qBAAM,CAAC;oBACP,OAAO,EAAE,CAAA;gBACV,CAAC;YACF,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}
+{"version":3,"file":"street.js","sourceRoot":"","sources":["../../lib/recipes/street.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,uBAAuB,CAAA;AAE/C,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAA;AAChD,OAAO,EAAE,aAAa,EAAE,UAAU,EAAqB,MAAM,mBAAmB,CAAA;AAChF,OAAO,EAAE,mBAAmB,EAAwB,MAAM,sBAAsB,CAAA;AAEhF;;;GAGG;AACH,MAAM,CAAC,MAAM,YAAY,GAAiB;IACzC,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,mFAAmF;IAChG,IAAI,EAAE,QAAQ;IACd,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,yBAAyB,EAAE,WAAW,EAAE,sCAAsC,EAAE,CAAC;IACnG,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,IAAI,CAAC,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,+CAA+C,CAAC,CAAA;QACjF,MAAM,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACjC,MAAM,sBAAsB,GAAG,IAAI,CAAC,eAAe,IAAI,IAAI,CAAA;QAC3D,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QAEf,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAClD,IAAI,EAAE,CAAA;YAEN,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBAC3E,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;gBAC5B,OAAO,EAAE,CAAA;gBAET,SAAQ;YACT,CAAC;YAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;gBACxC,MAAM,KAAK,GAAG,mBAAmB,CAAC,KAAwB,EAAE,EAAE,MAAM,EAAE,sBAAsB,EAAE,CAAC,CAAA;gBAE/F,IAAI,CAAC,KAAK;oBAAE,SAAQ;gBAEpB,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,EAAE,cAAc;oBACtB,SAAS,EAAE,cAAc,CAAC,cAAc,EAAE;wBACzC,QAAQ,EAAE,GAAG,KAAK,CAAC,QAAQ,IAAI,CAAC,EAAE;wBAClC,MAAM,EAAE,KAAK,CAAC,MAAM;wBACpB,QAAQ,EAAE,KAAK,CAAC,QAAQ;wBACxB,OAAO,EAAE,KAAK,CAAC,OAAO;qBACtB,CAAC;oBACF,cAAc,EAAE,OAAO;oBACvB,OAAO,EAAE,2DAA2D;iBACpE,EACD,eAAe,CACf,CAAA;gBAED,IAAI,EAAE,EAAE,CAAC;oBACR,OAAO,EAAE,CAAA;gBACV,CAAC;qBAAM,CAAC;oBACP,OAAO,EAAE,CAAA;gBACV,CAAC;YACF,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}

package/out/recipes/sub/venue/context.d.ts

@@ -0,0 +1,29 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ * @file Reads address context and derives street-side negatives for the sub-venue recipe.
+ */
+import type { PathBuilderLike } from "path-ts";
+import type { LocaleBaseTuple } from "#synthesizers/locale";
+/**
+ * Load the address skeletons every leg renders onto: GB / US / FR from the house-venue v3 tuples (the same 176,519 real
+ * rows the `synth-house-venue` slice is built from, so the two slices' address halves are drawn from one pool), DE and
+ * ES streamed out of OpenAddresses.
+ */
+export declare function loadContextTuples(tuplesPath: PathBuilderLike, seed: number): Promise<Map<string, LocaleBaseTuple[]>>;
+/**
+ * The street-side confound classes, mined from the leg's OWN address tuples.
+ *
+ * Real streets, not invented ones. The 176,519-row context pool carries 195 GB `hall` streets, 114 GB `gate` streets,
+ * 134 distinct GB `-gate` single tokens and a two-figure `<modifier> <designator>` population in both GB and US — small
+ * absolute numbers, but every one of them a street somebody lives on, which is the property an invented list cannot
+ * have.
+ */
+export interface StreetNegatives {
+    designator: LocaleBaseTuple[];
+    modifierDesignator: LocaleBaseTuple[];
+    gateSuffix: LocaleBaseTuple[];
+}
+export declare function buildStreetNegatives(context: readonly LocaleBaseTuple[], designatorPhrases: readonly string[], modifiers: readonly string[], country: string): StreetNegatives;
+//# sourceMappingURL=context.d.ts.map

package/out/recipes/sub/venue/context.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"context.d.ts","sourceRoot":"","sources":["../../../../lib/recipes/sub/venue/context.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAIH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAA;AAI9C,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAA;AAqB3D;;;;GAIG;AACH,wBAAsB,iBAAiB,CACtC,UAAU,EAAE,eAAe,EAC3B,IAAI,EAAE,MAAM,GACV,OAAO,CAAC,GAAG,CAAC,MAAM,EAAE,eAAe,EAAE,CAAC,CAAC,CA2CzC;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,eAAe;IAC/B,UAAU,EAAE,eAAe,EAAE,CAAA;IAC7B,kBAAkB,EAAE,eAAe,EAAE,CAAA;IACrC,UAAU,EAAE,eAAe,EAAE,CAAA;CAC7B;AAQD,wBAAgB,oBAAoB,CACnC,OAAO,EAAE,SAAS,eAAe,EAAE,EACnC,iBAAiB,EAAE,SAAS,MAAM,EAAE,EACpC,SAAS,EAAE,SAAS,MAAM,EAAE,EAC5B,OAAO,EAAE,MAAM,GACb,eAAe,CAkCjB"}

package/out/recipes/sub/venue/context.js

@@ -0,0 +1,106 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ * @file Reads address context and derives street-side negatives for the sub-venue recipe.
+ */
+import { dataRootPath } from "@mailwoman/core/data-root";
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
+import { readTuples as readLocaleTuples } from "#international/recipes/locale";
+import { readTuples as readSliceTuples } from "#recipes/scaffold";
+//#region Address context
+/**
+ * DE + ES address context. Both read through {@link readLocaleTuples}, the `locale` recipe's own streaming + reservoir
+ * reader, so the CSV handling (quoted fields, CRLF, city-noise cleaning, the DE per-part region fallback) has exactly
+ * one implementation.
+ *
+ * DE reads `europe.zip`'s two members rather than `oa-cache/de__*.zip`: the cached per-state zips the `locale` recipe
+ * names are not materialized on this host, and the archive members are byte-identical to OA's current run (verified in
+ * `corpus/AGENTS.md`'s "a file's mtime is not its data's vintage" note).
+ */
+const CONTEXT_PARTS = {
+    DE: [
+        { zip: dataRootPath("openaddresses", "europe.zip"), csv: "de/berlin.csv", region: "Berlin" },
+        { zip: dataRootPath("openaddresses", "europe.zip"), csv: "de/sn/statewide.csv", region: "Sachsen" },
+    ],
+    ES: [{ path: dataRootPath("openaddresses", "extracted", "es", "countrywide.csv") }],
+};
+/**
+ * Load the address skeletons every leg renders onto: GB / US / FR from the house-venue v3 tuples (the same 176,519 real
+ * rows the `synth-house-venue` slice is built from, so the two slices' address halves are drawn from one pool), DE and
+ * ES streamed out of OpenAddresses.
+ */
+export async function loadContextTuples(tuplesPath, seed) {
+    const byCountry = new Map();
+    for await (const tuple of readSliceTuples(tuplesPath)) {
+        const country = String(tuple.country ?? "");
+        if (!country || !tuple.locality || !tuple.street)
+            continue;
+        const mapped = {
+            house_number: String(tuple.houseNumber ?? tuple.house_number ?? ""),
+            street: String(tuple.street),
+            locality: String(tuple.locality),
+            region: String(tuple.region ?? ""),
+            postcode: String(tuple.postcode ?? ""),
+        };
+        const list = byCountry.get(country);
+        if (list) {
+            list.push(mapped);
+        }
+        else {
+            byCountry.set(country, [mapped]);
+        }
+    }
+    for (const [country, parts] of Object.entries(CONTEXT_PARTS)) {
+        if (byCountry.has(country))
+            continue;
+        const pooled = [];
+        for (const [index, part] of parts.entries()) {
+            // A dedicated stream PRNG, seeded per part, so the input sample is reproducible without
+            // perturbing the emit loop's draws (the `locale` recipe's rule, kept).
+            const streamRandom = makeMulberry32(seed + index);
+            for (const tuple of await readLocaleTuples(part, streamRandom)) {
+                pooled.push(tuple);
+            }
+        }
+        byCountry.set(country, pooled);
+    }
+    return byCountry;
+}
+/**
+ * Shortest token that can carry a `-gate` street suffix and still be a NAME rather than the bare word: `gate` itself is
+ * four characters, so the class starts at five (`Highgate`, `Moorgate`, `Stonegate`).
+ */
+const MIN_GATE_SUFFIX_TOKEN_LENGTH = 5;
+export function buildStreetNegatives(context, designatorPhrases, modifiers, country) {
+    const designatorSet = new Set(designatorPhrases);
+    const modifierSet = new Set(modifiers);
+    const designator = [];
+    const modifierDesignator = [];
+    const gateSuffix = [];
+    for (const tuple of context) {
+        const tokens = tuple.street.toLowerCase().match(/[\p{L}]+/gu) ?? [];
+        let isDesignator = false;
+        let isPair = false;
+        for (const [index, token] of tokens.entries()) {
+            if (designatorSet.has(token)) {
+                isDesignator = true;
+            }
+            if (index + 1 < tokens.length && modifierSet.has(token) && designatorSet.has(tokens[index + 1])) {
+                isPair = true;
+            }
+            if (country === "GB" && token.length >= MIN_GATE_SUFFIX_TOKEN_LENGTH && token.endsWith("gate")) {
+                gateSuffix.push(tuple);
+            }
+        }
+        if (isPair) {
+            modifierDesignator.push(tuple);
+        }
+        else if (isDesignator) {
+            designator.push(tuple);
+        }
+    }
+    return { designator, modifierDesignator, gateSuffix };
+}
+//#endregion
+//# sourceMappingURL=context.js.map

package/out/recipes/sub/venue/context.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"context.js","sourceRoot":"","sources":["../../../../lib/recipes/sub/venue/context.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAA;AACxD,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AAGpE,OAAO,EAAE,UAAU,IAAI,gBAAgB,EAAmB,MAAM,+BAA+B,CAAA;AAC/F,OAAO,EAAE,UAAU,IAAI,eAAe,EAAE,MAAM,mBAAmB,CAAA;AAGjE,yBAAyB;AAEzB;;;;;;;;GAQG;AACH,MAAM,aAAa,GAAoD;IACtE,EAAE,EAAE;QACH,EAAE,GAAG,EAAE,YAAY,CAAC,eAAe,EAAE,YAAY,CAAC,EAAE,GAAG,EAAE,eAAe,EAAE,MAAM,EAAE,QAAQ,EAAE;QAC5F,EAAE,GAAG,EAAE,YAAY,CAAC,eAAe,EAAE,YAAY,CAAC,EAAE,GAAG,EAAE,qBAAqB,EAAE,MAAM,EAAE,SAAS,EAAE;KACnG;IACD,EAAE,EAAE,CAAC,EAAE,IAAI,EAAE,YAAY,CAAC,eAAe,EAAE,WAAW,EAAE,IAAI,EAAE,iBAAiB,CAAC,EAAE,CAAC;CACnF,CAAA;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACtC,UAA2B,EAC3B,IAAY;IAEZ,MAAM,SAAS,GAAG,IAAI,GAAG,EAA6B,CAAA;IAEtD,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,eAAe,CAAC,UAAU,CAAC,EAAE,CAAC;QACvD,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAA;QAE3C,IAAI,CAAC,OAAO,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,MAAM;YAAE,SAAQ;QAE1D,MAAM,MAAM,GAAoB;YAC/B,YAAY,EAAE,MAAM,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;YACnE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC;YAC5B,QAAQ,EAAE,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC;YAChC,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC;YAClC,QAAQ,EAAE,MAAM,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,CAAC;SACtC,CAAA;QAED,MAAM,IAAI,GAAG,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;QAEnC,IAAI,IAAI,EAAE,CAAC;YACV,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QAClB,CAAC;aAAM,CAAC;YACP,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,CAAC,MAAM,CAAC,CAAC,CAAA;QACjC,CAAC;IACF,CAAC;IAED,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,aAAa,CAAC,EAAE,CAAC;QAC9D,IAAI,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,SAAQ;QACpC,MAAM,MAAM,GAAsB,EAAE,CAAA;QAEpC,KAAK,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC;YAC7C,wFAAwF;YACxF,uEAAuE;YACvE,MAAM,YAAY,GAAG,cAAc,CAAC,IAAI,GAAG,KAAK,CAAC,CAAA;YAEjD,KAAK,MAAM,KAAK,IAAI,MAAM,gBAAgB,CAAC,IAAI,EAAE,YAAY,CAAC,EAAE,CAAC;gBAChE,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;YACnB,CAAC;QACF,CAAC;QAED,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAA;IAC/B,CAAC;IAED,OAAO,SAAS,CAAA;AACjB,CAAC;AAgBD;;;GAGG;AACH,MAAM,4BAA4B,GAAG,CAAC,CAAA;AAEtC,MAAM,UAAU,oBAAoB,CACnC,OAAmC,EACnC,iBAAoC,EACpC,SAA4B,EAC5B,OAAe;IAEf,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,iBAAiB,CAAC,CAAA;IAChD,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,CAAA;IACtC,MAAM,UAAU,GAAsB,EAAE,CAAA;IACxC,MAAM,kBAAkB,GAAsB,EAAE,CAAA;IAChD,MAAM,UAAU,GAAsB,EAAE,CAAA;IAExC,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,CAAA;QACnE,IAAI,YAAY,GAAG,KAAK,CAAA;QACxB,IAAI,MAAM,GAAG,KAAK,CAAA;QAElB,KAAK,MAAM,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,EAAE,CAAC;YAC/C,IAAI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC9B,YAAY,GAAG,IAAI,CAAA;YACpB,CAAC;YAED,IAAI,KAAK,GAAG,CAAC,GAAG,MAAM,CAAC,MAAM,IAAI,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,aAAa,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAE,CAAC,EAAE,CAAC;gBAClG,MAAM,GAAG,IAAI,CAAA;YACd,CAAC;YAED,IAAI,OAAO,KAAK,IAAI,IAAI,KAAK,CAAC,MAAM,IAAI,4BAA4B,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;gBAChG,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;YACvB,CAAC;QACF,CAAC;QAED,IAAI,MAAM,EAAE,CAAC;YACZ,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;QAC/B,CAAC;aAAM,IAAI,YAAY,EAAE,CAAC;YACzB,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;QACvB,CAAC;IACF,CAAC;IAED,OAAO,EAAE,UAAU,EAAE,kBAAkB,EAAE,UAAU,EAAE,CAAA;AACtD,CAAC;AAED,YAAY"}

package/out/recipes/sub/venue/index.d.ts

@@ -0,0 +1,270 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `sub-venue` slice recipe (#35 step 4) — teach the `unit` tag the venue-INTERIOR shapes it was
+ *   never taught, so the modifier+designator class wins at the shipped `venueStructureBiasScale` of
+ *   6.0 instead of needing ~11 nats. `docs/engineering/sub-venue-corpus-task.mdx` is the spec; the
+ *   vocabulary is `corpus/data/sub-venue-lexicon.json` (v0.2.0) and the curation ledger is
+ *   `corpus/src/tools/sub-venue-promotions.ts`. The READ half — promotions, identifier
+ *   distributions, name pools — is `sub-venue-sources.ts`; this file renders lines and emits rows.
+ *
+ *   ── WHY THIS SYNTHESIZES RATHER THAN HARVESTS ────────────────────────────────────────────────────
+ *   Wave 1's lesson, and the reason the spec's "get real data first" instruction is honoured in a
+ *   shape it did not anticipate: the attested SURFACE STRINGS are thin. 87 GB features attest
+ *   `terminal`, 29 attest `wing`, 4 attest `concourse` (3 of which are a street called CONCOURSE WAY).
+ *   You cannot train a tag on 29 strings. What the five extracts DO carry at volume is the three
+ *   things a generator needs — 45,000+ real venue names across four countries, a per-region
+ *   identifier DISTRIBUTION measured over 2,868 gate/terminal/campus refs, and the confound
+ *   population that becomes the negatives. So the bulk is `designator × per-region-identifier ×
+ *   modifier` sampled per locale, and the attested strings ride along as seasoning
+ *   ({@link ATTESTED_FRACTION}) rather than as the corpus.
+ *
+ *   ── THE PER-REGION IDENTIFIER RULE, AND WHY IT IS NOT COSMETIC ───────────────────────────────────
+ *   `Gate A12` is a rendering, not a string anyone wrote down: all but 13 of Great Britain's 658
+ *   `aeroway=gate` features are unnamed and carry only a `ref`. The lexicon therefore ships a
+ *   distribution, and it differs by country far more than the shared English vocabulary suggests —
+ *   GB gates are 71% bare digits and JP 89%, FR and DE are ~60% letter-digit (`A37`, `B05`), and ES
+ *   gives a THIRD of its gates a range (`B18-B20`), which no other country does at that rate. A
+ *   generator that samples Great Britain's shape into a Spanish line produces a plausible string that
+ *   is wrong about Spain, so every leg samples its own region.
+ *
+ *   ── ONLY PROMOTED (designator, locale) PAIRS PRODUCE POSITIVES ───────────────────────────────────
+ *   A promotion names a designator, a phrase AND a locale, because the same token is a designator in
+ *   one language and a disaster in another: `hall` is 0-of-3,273 in Great Britain and 35-of-40 in
+ *   France; `wing` is 23-of-29 in Great Britain and 4-of-3,358 in the United States. A REJECTED pair
+ *   generates NEGATIVES in that locale instead — en-US `wing` rows are Red Wing, not units.
+ *
+ *   `shape: "identifier-required"` is honoured as the ledger's docstring demands: de-DE `halle` is
+ *   emitted only as `Halle <identifier>`, never bare and never after a modifier, because its 168-hit
+ *   confound includes the CITY Halle (Saale) and only the identifier-bearing shape separates them.
+ *   {@link buildSubVenueForm} enforces it and `sub-venue.test.ts` pins it.
+ *
+ *   ── LABELS ───────────────────────────────────────────────────────────────────────────────────────
+ *   Sub-venue is `unit`; the container is `venue`. That is the spec's wording and it invents nothing:
+ *   `block` / `sub_block` exist in the `ComponentTag` union but are JP-char-model-only and outside
+ *   `ACTIVE_TAGS` (STAGE3), so they are not reachable from a Latin slice.
+ *
+ *   ── WHAT IS DELIBERATELY NOT HERE ────────────────────────────────────────────────────────────────
+ *   1. **A modifier+designator form outside English.** `VENUE_STRUCTURE_MODIFIERS` is an English
+ *        list, and the extracts say the localized modifier surfaces do not exist to copy: `aile` in
+ *        France is 0 hits, `ala` in Spain 0, `flügel` in Germany 0. Generating `Terminal Sud` would
+ *        be inventing a vocabulary with no confound board behind it, which is the exact failure the
+ *        promotion ledger exists to prevent. Non-English legs get designator+identifier only.
+ *   2. **A ja-JP leg.** `ターミナル` is promoted (1,213 real of 1,215). The one non-Latin
+ *        surface the task named, but Japanese addresses train through `build_jp_slice.py` against the
+ *        `stage3-jp` 47-label head, where the interior tags are `block`/`sub_block`/`building_number`
+ *        — a different model, a different label set, and a different builder. A katakana `unit` row
+ *        in this (Latin) feed would in any case be dropped by `country_weights`, which carries no
+ *        `JP` key. The JP extract is harvested and ready; the leg belongs to the JP slice.
+ */
+import type { ComponentTag } from "@mailwoman/codex/component";
+import { type CorpusRecipe } from "#recipes/scaffold";
+import { type IdentifierModel, type PromotedSurface } from "#recipes/sub/venue/sources";
+import type { LocaleBaseTuple } from "#synthesizers/locale";
+export * from "#recipes/sub/venue/sources";
+export * from "#recipes/sub/venue/context";
+/**
+ * One locale's leg of the slice.
+ *
+ * `positiveShare` / `negativeShare` are relative weights, normalized at run time — they do not have to sum to 1.
+ */
+export interface SubVenueLeg {
+    locale: string;
+    country: string;
+    /**
+     * ISO 3166-1 alpha-2 key into the lexicon's `identifierShapes` AND the extract filename. The two axes are the same
+     * axis: a distribution is measured in a region's own extract.
+     */
+    region: string;
+    /**
+     * Extract filename under `--extracts-dir`. Absent = no OSM extract for this leg (en-US), which then draws its venue
+     * and confound pools from `poi.db` instead.
+     */
+    extract?: string;
+    /**
+     * May this leg use the English `<modifier> <designator>` grammar? See the module docstring's exclusion 1.
+     */
+    english: boolean;
+    positiveShare: number;
+    negativeShare: number;
+    /**
+     * Ca-ES only — keep context tuples whose postcode starts with one of these. Catalan-language territories by postal
+     * prefix (07 Illes Balears, 08 Barcelona, 17 Girona, 25 Lleida, 43 Tarragona) rather than by a REGION string, whose
+     * spelling in the OA export is not something to guess at.
+     */
+    postcodePrefixes?: readonly string[];
+}
+/**
+ * The legs, and the numbers behind the shares.
+ *
+ * En-GB and en-US carry the most because they are where the eval board lives (28 of the 30 confound rows are GB or US
+ * addresses) and because the English shipped vocabulary is the only one with a modifier grammar — the target class.
+ * fr-FR, de-DE and es-ES exist because the ledger promoted surfaces there (169, 19+32 and 190 real hits respectively)
+ * and a slice that skipped them would leave every non-English promotion untrained. ca-ES is small on purpose: its
+ * promotion is 15 hits and its line differs from es-ES only in the Catalan street vocabulary the postal-prefix filter
+ * selects for.
+ *
+ * The negative shares invert that ordering where the confound mass does. en-US carries the largest negative share
+ * because its confound population is the largest measured anywhere in the ledger — 3,354 `wing` (Red Wing boots 676,
+ * chicken wings 759), 2,330 `pier` (Pier 1 Imports, which IS the designator+identifier shape), 27,081 `hall`.
+ */
+export declare const SUBVENUE_LEGS: readonly SubVenueLeg[];
+/**
+ * En-US has no OSM extract, so its identifier distribution has to be borrowed. GB is the borrow, and the leg's `region`
+ * says so literally rather than in a comment: the two English-speaking aviation systems number their gates the same way
+ * (GB 71% bare digit) and poi.db — the only US source in reach — carries names, not refs, so it cannot supply a
+ * distribution of its own. Recorded here because it is the one place a leg's `region` is not its own country.
+ */
+export declare const US_IDENTIFIER_REGION_BORROWED_FROM = "GB";
+/**
+ * The row count this slice is built at, and the arithmetic behind it. `--count` overrides; this is the number to use
+ * absent a reason.
+ *
+ * The training sampler (`corpus-python/src/mailwoman_train/data_loader.py`, `_raw_row_stream`) draws SOURCES from a
+ * multinomial over `source_weights` and yields the next row from that source's iterator. Two consequences set the
+ * size:
+ *
+ * 1. A source's share of an epoch is `weight / Σweights`, independent of how many rows it has.
+ * 2. **A source that exhausts is DELETED from the multinomial** — there is no cycling. Under-size the slice and its
+ *    nominal dose is fiction for the rest of the epoch.
+ *
+ * Measured against the shipped `v4.1.0-gb-venue-l1e4-2k` weight table: 33 sources summing to 144.5. At the dose the B11
+ * GB-venue exercise settled on for a hard rare class — 12.0, the value `synth-fr-bare-street`, `synth-si-bare-village`,
+ * `synth-cz-pcfirst-preposition`, `synth-fr-fragment` and `synth-no-fragment` all carry — the share is `12 / 156.5 =
+ * 7.67%`, and `train_rows_per_epoch` is 1,000,000. So the epoch draws **76,677 rows** from this slice, and anything
+ * smaller runs dry mid-epoch. 120,000 clears that with room for a config that drops a source or raises the dose. (For
+ * contrast: `synth-fr-bare-street` is 10,803 rows at dose 12.0, so it exhausts 14% into its own nominal share every
+ * epoch — a precedent for the dose, not for the size.)
+ */
+export declare const RECOMMENDED_ROW_COUNT = 120000;
+/**
+ * Surfaces reserved by `mailwoman/eval-harness/fixtures/venue-structure-confounds.jsonl` — the 30-row board this slice
+ * has to hold. A row containing any of these is DROPPED and counted in `contaminated`.
+ *
+ * The `--exclude-surfaces` precedent from `fr-fragment` / `no-fragment`, applied by hand rather than by file because
+ * the board lives in `mailwoman/` and `@mailwoman/corpus` cannot reach across that workspace boundary at run time. Keep
+ * it in sync when the board grows; a slice that trains on its own eval set measures memorization.
+ *
+ * Note what this costs and why it is still right: reserving `east gate` / `west gate` removes the two GB surfaces the
+ * board uses for its `modifier-designator-street` class, so the slice teaches that class from the OTHER real ones its
+ * sources carry (`North Gate`, `South Gate`, `East Hall`, `West Hall`, `Lower Hall`, `East Campus`, …). The class is
+ * taught; the board's own strings are not.
+ */
+export declare const BOARD_RESERVED_SURFACES: readonly string[];
+/**
+ * Does this row's text collide with a board-reserved surface?
+ */
+export declare function isBoardReserved(raw: string): boolean;
+/**
+ * One labelled piece of the line. Pieces inside a group are space-joined; groups are joined by the register's
+ * separator.
+ */
+interface Piece {
+    text: string;
+    tag?: ComponentTag;
+}
+type Group = Piece[];
+/**
+ * Surface register. Every eval in this repo gets a lowercase leg because lowercase is the register users type — Google
+ * Maps taught them — so every slice has to carry one.
+ */
+declare const Register: {
+    readonly Canonical: "canonical";
+    readonly CommaFree: "comma-free";
+    readonly Lower: "lower";
+    readonly Upper: "upper";
+};
+type Register = (typeof Register)[keyof typeof Register];
+/**
+ * Join groups into `raw` + `components`, applying the register to BOTH so alignment still finds every value.
+ */
+export declare function renderGroups(groups: Group[], register: Register): {
+    raw: string;
+    components: Partial<Record<ComponentTag, string>>;
+};
+/**
+ * The street + tail groups for a country, in that country's own order.
+ *
+ * DE/ES/FR put the postcode before the locality and DE/ES put the house number after the street; GB and US keep the
+ * anglophone order and US carries a region. These are the same orders `synthesizers/locale.ts` renders, restated here
+ * because this recipe assembles its groups piece-by-piece (it has to, to place a sub-venue group in front of them).
+ */
+export declare function addressGroups(country: string, tuple: LocaleBaseTuple, withStreet: boolean): Group[];
+/**
+ * A sub-venue string plus how it was made, for the composition report.
+ */
+export interface SubVenueForm {
+    text: string;
+    form: "designator-identifier" | "modifier-designator" | "attested";
+    designatorID: string;
+}
+/**
+ * Build one sub-venue string for a leg.
+ *
+ * The `identifier-required` guard is here and not at the call site on purpose: it is the one rule in this file that a
+ * refactor must not be able to route around. A promotion carrying `shape: "identifier-required"` can only ever leave
+ * this function as `<Phrase> <identifier>`, and returns `null` rather than a bare or modified form.
+ */
+export declare function buildSubVenueForm(leg: SubVenueLeg, promoted: readonly PromotedSurface[], model: IdentifierModel, modifiers: readonly string[], attested: readonly string[], random: () => number): SubVenueForm | null;
+/**
+ * Alias kept for readers of the arc's earlier drafts.
+ *
+ * @deprecated Use {@link buildSubVenueForm}.
+ */
+export declare const buildPositiveForms: typeof buildSubVenueForm;
+/**
+ * Negative classes, named so the composition report can count them and a failure can be attributed.
+ */
+export declare const NegativeClass: {
+    /**
+     * A locale-REJECTED surface in the venue slot: Red Wing Shoes, Village Hall, Porte de Champerret.
+     */
+    readonly RejectedVenue: "rejected-venue";
+    /**
+     * The designator inside a longer proper name, whole string tagged `venue`: Lochaline Ferry Terminal.
+     */
+    readonly LongerName: "longer-name";
+    /**
+     * A real street whose name carries a designator token: Pier Road, Egg Hall, Orchard Gate.
+     */
+    readonly DesignatorStreet: "designator-street";
+    /**
+     * A real street of the `<modifier> <designator>` shape — the class that would otherwise be read as a sub-venue.
+     */
+    readonly ModifierDesignatorStreet: "modifier-designator-street";
+    /**
+     * A GB single-token `-gate` street: Eastgate, Southgate, Moorgate, Stonegate.
+     */
+    readonly GateSuffixStreet: "gate-suffix-street";
+    /**
+     * A PROMOTED phrase outside the shape its promotion covers — `Halle Rosengarten`, `PHOENIX Halle`. The other half of
+     * an `identifier-required` ruling; see `LegPools.unpromotedShapes`.
+     */
+    readonly UnpromotedShape: "unpromoted-shape";
+};
+export type NegativeClass = (typeof NegativeClass)[keyof typeof NegativeClass];
+/**
+ * Per-leg composition tallies the build prints and the report quotes.
+ */
+export interface SubVenueLegStats {
+    locale: string;
+    positives: number;
+    negatives: number;
+    byForm: Record<string, number>;
+    byDesignator: Record<string, number>;
+    byNegativeClass: Record<string, number>;
+    byRegister: Record<string, number>;
+    poolSizes: Record<string, number>;
+}
+/**
+ * Split `total` across `shares` (which need not sum to 1), largest-remainder so the parts sum exactly.
+ */
+export declare function allocate(total: number, shares: readonly number[]): number[];
+/**
+ * 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 subVenueRecipe: CorpusRecipe;
+//# sourceMappingURL=index.d.ts.map

package/out/recipes/sub/venue/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../lib/recipes/sub/venue/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4DG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,4BAA4B,CAAA;AAM9D,OAAO,EAAiB,KAAK,YAAY,EAAE,MAAM,mBAAmB,CAAA;AAEpE,OAAO,EAIN,KAAK,eAAe,EAIpB,KAAK,eAAe,EAQpB,MAAM,4BAA4B,CAAA;AACnC,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAA;AAK3D,cAAc,4BAA4B,CAAA;AAC1C,cAAc,4BAA4B,CAAA;AAI1C;;;;GAIG;AACH,MAAM,WAAW,WAAW;IAC3B,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,MAAM,CAAA;IACf;;;OAGG;IACH,MAAM,EAAE,MAAM,CAAA;IACd;;;OAGG;IACH,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;;OAEG;IACH,OAAO,EAAE,OAAO,CAAA;IAChB,aAAa,EAAE,MAAM,CAAA;IACrB,aAAa,EAAE,MAAM,CAAA;IACrB;;;;OAIG;IACH,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;CACpC;AAED;;;;;;;;;;;;;GAaG;AACH,eAAO,MAAM,aAAa,EAAE,SAAS,WAAW,EAgD/C,CAAA;AAED;;;;;GAKG;AACH,eAAO,MAAM,kCAAkC,OAAO,CAAA;AAMtD;;;;;;;;;;;;;;;;;;;GAmBG;AACH,eAAO,MAAM,qBAAqB,SAAU,CAAA;AAgC5C;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,uBAAuB,EAAE,SAAS,MAAM,EAkCpD,CAAA;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAIpD;AAMD;;;GAGG;AACH,UAAU,KAAK;IACd,IAAI,EAAE,MAAM,CAAA;IACZ,GAAG,CAAC,EAAE,YAAY,CAAA;CAClB;AAED,KAAK,KAAK,GAAG,KAAK,EAAE,CAAA;AAEpB;;;GAGG;AACH,QAAA,MAAM,QAAQ;aACb,SAAS,EAAE,WAAW;aACtB,SAAS,EAAE,YAAY;aACvB,KAAK,EAAE,OAAO;aACd,KAAK,EAAE,OAAO;CACL,CAAA;AAEV,KAAK,QAAQ,GAAG,CAAC,OAAO,QAAQ,CAAC,CAAC,MAAM,OAAO,QAAQ,CAAC,CAAA;AAmBxD;;GAEG;AACH,wBAAgB,YAAY,CAC3B,MAAM,EAAE,KAAK,EAAE,EACf,QAAQ,EAAE,QAAQ,GAChB;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,OAAO,CAAC,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC,CAAA;CAAE,CA2BpE;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,eAAe,EAAE,UAAU,EAAE,OAAO,GAAG,KAAK,EAAE,CAwDnG;AAMD;;GAEG;AACH,MAAM,WAAW,YAAY;IAC5B,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,uBAAuB,GAAG,qBAAqB,GAAG,UAAU,CAAA;IAClE,YAAY,EAAE,MAAM,CAAA;CACpB;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAChC,GAAG,EAAE,WAAW,EAChB,QAAQ,EAAE,SAAS,eAAe,EAAE,EACpC,KAAK,EAAE,eAAe,EACtB,SAAS,EAAE,SAAS,MAAM,EAAE,EAC5B,QAAQ,EAAE,SAAS,MAAM,EAAE,EAC3B,MAAM,EAAE,MAAM,MAAM,GAClB,YAAY,GAAG,IAAI,CAiCrB;AAED;;;;GAIG;AACH,eAAO,MAAM,kBAAkB,0BAAoB,CAAA;AAMnD;;GAEG;AACH,eAAO,MAAM,aAAa;IACzB;;OAEG;aACH,aAAa,EAAE,gBAAgB;IAC/B;;OAEG;aACH,UAAU,EAAE,aAAa;IACzB;;OAEG;aACH,gBAAgB,EAAE,mBAAmB;IACrC;;OAEG;aACH,wBAAwB,EAAE,4BAA4B;IACtD;;OAEG;aACH,gBAAgB,EAAE,oBAAoB;IACtC;;;OAGG;aACH,eAAe,EAAE,kBAAkB;CAC1B,CAAA;AAEV,MAAM,MAAM,aAAa,GAAG,CAAC,OAAO,aAAa,CAAC,CAAC,MAAM,OAAO,aAAa,CAAC,CAAA;AA+B9E;;GAEG;AACH,MAAM,WAAW,gBAAgB;IAChC,MAAM,EAAE,MAAM,CAAA;IACd,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC9B,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACpC,eAAe,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACvC,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAClC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CACjC;AA6MD;;GAEG;AACH,wBAAgB,QAAQ,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,EAAE,CAiB3E;AA6DD;;;GAGG;AACH,eAAO,MAAM,cAAc,EAAE,YAiG5B,CAAA"}