@mailwoman/corpus 10.0.0 → 10.1.0

package/out/recipes/bare/country.js

@@ -1,32 +1,22 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   `bare-country` — the country name as the WHOLE query. Addresses almost never consist of one bare
- *   country name, so the class is out of the model's training distribution and the tag is a coin
- *   flip: measured on the 2026-08-13 panel, France/Germany/United States/New Zealand parsed
- *   `country` while Japan/China/Nigeria/Australia/Deutschland parsed `locality`. The retrieval-side
- *   bare-country race (#1651) covers the answer either way; this slice closes the PARSE half so the
- *   tag itself is right.
- *
- *   Source is the codex country table — every ISO canonical English name plus the curated surface
- *   forms (endonyms and long-form names). The 2–3 letter CODES are deliberately excluded: a bare
- *   `JP` or `GER` is an abbreviation register with its own ambiguity surface (`IN`, `DE`, `TO` are
- *   English words), and teaching it here would be an unmeasured claim. Case augmentation is the
- *   loader's job, as everywhere.
- */
 import { COUNTRY_SURFACE_FORMS, CountryNames, matchCountry } from "@mailwoman/codex/country";
 import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
-import { alignAndWrite, sliceSourceID } from "#recipes/scaffold";
+import { alignAndWrite, recipeSourceID } from "#recipes/scaffold";
+import { defaultRecipeSource } from "#recipes/sources";
+import { SourceRegister } from "#registers";
+import { SurfaceOrigin } from "#types";
 /**
- * Surfaces shorter than this are the code register (`JP`, `GER`), not a name — excluded (see the module doc).
+ * Resolved once, because `source_id` includes the source as its prefix and a pair
+ * that disagreed would name a source no row of this output declares.
  */
+const BARE_COUNTRY_SOURCE = defaultRecipeSource("synth-bare-country");
+const BARE_COUNTRY_PROVENANCE = {
+    register: SourceRegister.Codex,
+    surface: SurfaceOrigin.Attested,
+    // The codex table is the record.
+    // Its rows have no `source_id` of their own.
+    baseSourceID: null,
+};
 const MIN_NAME_LENGTH = 4;
-/**
- * Every (surface, iso2) pair the slice emits: the ISO canonical names plus each country's curated surface forms,
- * deduplicated on the surface string (a form shared across countries — none known — would keep its first bearer).
- */
 function* bareCountrySurfaces() {
     const seen = new Set();
     for (const [iso2, forms] of Object.entries(COUNTRY_SURFACE_FORMS)) {
@@ -48,15 +38,14 @@ function* bareCountrySurfaces() {
     }
 }
 /**
- * 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.
+ * Generates queries that consist only of a country name, drawn from ISO names
+ * and curated surface forms of at least four characters, with no country codes.
  */
 export const bareCountryRecipe = {
     name: "bare-country",
-    description: "The country name as the whole query (#1651 parse half): ISO names + curated endonyms, no codes",
+    description: "Emits a country name as the whole query, drawn from ISO names and curated endonyms but never from country codes",
     mode: "generate",
     async run(opts, write) {
-        // Seeded for parity with the other recipes; the surfaces themselves drive the content.
         makeMulberry32(opts.seed);
         let read = 0;
         let emitted = 0;
@@ -64,18 +53,18 @@ export const bareCountryRecipe = {
         for (const { surface, iso2 } of bareCountrySurfaces()) {
             read++;
             const components = { country: surface };
-            const source_id = sliceSourceID("synth-bare-country", { country: surface, cc: iso2, v: String(read) });
+            const source_id = recipeSourceID(BARE_COUNTRY_SOURCE, { country: surface, cc: iso2, v: String(read) });
             const canonical = {
                 raw: surface,
                 components,
                 country: iso2,
                 locale: "und",
-                source: "synth-bare-country",
+                source: BARE_COUNTRY_SOURCE,
                 source_id,
                 corpus_version: "0.11.0",
                 license: "Synthetic — bare-country; surfaces from the codex country table (ISO 3166 + curated forms)",
             };
-            if (alignAndWrite(write, canonical, "bare-country")) {
+            if (alignAndWrite(write, canonical, "bare-country", BARE_COUNTRY_PROVENANCE)) {
                 emitted++;
             }
             else {

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

@@ -1 +1 @@
-{"version":3,"file":"country.js","sourceRoot":"","sources":["../../../lib/recipes/bare/country.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,qBAAqB,EAAE,YAAY,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAA;AAC5F,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AAEpE,OAAO,EAAE,aAAa,EAAqB,aAAa,EAAE,MAAM,mBAAmB,CAAA;AACnF;;GAEG;AACH,MAAM,eAAe,GAAG,CAAC,CAAA;AAEzB;;;GAGG;AACH,QAAQ,CAAC,CAAC,mBAAmB;IAC5B,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAE9B,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,qBAAqB,CAAC,EAAE,CAAC;QACnE,KAAK,MAAM,OAAO,IAAI,KAAK,EAAE,CAAC;YAC7B,IAAI,OAAO,CAAC,MAAM,GAAG,eAAe,IAAI,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC;gBAAE,SAAQ;YACnE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;YACjB,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,CAAA;QACxB,CAAC;IACF,CAAC;IAED,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QACjC,IAAI,IAAI,CAAC,MAAM,GAAG,eAAe,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,SAAQ;QAC7D,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,CAAA;QAElC,IAAI,CAAC,OAAO;YAAE,SAAQ;QACtB,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;QACd,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAA;IAC5C,CAAC;AACF,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAiB;IAC9C,IAAI,EAAE,cAAc;IACpB,WAAW,EAAE,gGAAgG;IAC7G,IAAI,EAAE,UAAU;IAChB,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,uFAAuF;QACvF,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACzB,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QAEf,KAAK,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,mBAAmB,EAAE,EAAE,CAAC;YACvD,IAAI,EAAE,CAAA;YAEN,MAAM,UAAU,GAA2B,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;YAC/D,MAAM,SAAS,GAAG,aAAa,CAAC,oBAAoB,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;YAEtG,MAAM,SAAS,GAAG;gBACjB,GAAG,EAAE,OAAO;gBACZ,UAAU;gBACV,OAAO,EAAE,IAAI;gBACb,MAAM,EAAE,KAAK;gBACb,MAAM,EAAE,oBAAoB;gBAC5B,SAAS;gBACT,cAAc,EAAE,QAAQ;gBACxB,OAAO,EAAE,4FAA4F;aACrG,CAAA;YAED,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,EAAE,cAAc,CAAC,EAAE,CAAC;gBACrD,OAAO,EAAE,CAAA;YACV,CAAC;iBAAM,CAAC;gBACP,OAAO,EAAE,CAAA;YACV,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}
+{"version":3,"file":"country.js","sourceRoot":"","sources":["../../../lib/recipes/bare/country.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,YAAY,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAA;AAC5F,OAAO,EAAE,UAAU,IAAI,cAAc,EAAE,MAAM,uBAAuB,CAAA;AAEpE,OAAO,EAAE,aAAa,EAAqB,cAAc,EAAE,MAAM,mBAAmB,CAAA;AACpF,OAAO,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAA;AACtD,OAAO,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAC3C,OAAO,EAAE,aAAa,EAAE,MAAM,QAAQ,CAAA;AAEtC;;;GAGG;AACH,MAAM,mBAAmB,GAAG,mBAAmB,CAAC,oBAAoB,CAAC,CAAA;AAErE,MAAM,uBAAuB,GAAG;IAC/B,QAAQ,EAAE,cAAc,CAAC,KAAK;IAC9B,OAAO,EAAE,aAAa,CAAC,QAAQ;IAC/B,iCAAiC;IACjC,6CAA6C;IAC7C,YAAY,EAAE,IAAI;CAClB,CAAA;AAED,MAAM,eAAe,GAAG,CAAC,CAAA;AAEzB,QAAQ,CAAC,CAAC,mBAAmB;IAC5B,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAE9B,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,qBAAqB,CAAC,EAAE,CAAC;QACnE,KAAK,MAAM,OAAO,IAAI,KAAK,EAAE,CAAC;YAC7B,IAAI,OAAO,CAAC,MAAM,GAAG,eAAe,IAAI,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC;gBAAE,SAAQ;YACnE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;YACjB,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,CAAA;QACxB,CAAC;IACF,CAAC;IAED,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QACjC,IAAI,IAAI,CAAC,MAAM,GAAG,eAAe,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,SAAQ;QAC7D,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,CAAA;QAElC,IAAI,CAAC,OAAO;YAAE,SAAQ;QACtB,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;QACd,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAA;IAC5C,CAAC;AACF,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAiB;IAC9C,IAAI,EAAE,cAAc;IACpB,WAAW,EACV,iHAAiH;IAClH,IAAI,EAAE,UAAU;IAChB,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACzB,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QAEf,KAAK,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,mBAAmB,EAAE,EAAE,CAAC;YACvD,IAAI,EAAE,CAAA;YAEN,MAAM,UAAU,GAA2B,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;YAC/D,MAAM,SAAS,GAAG,cAAc,CAAC,mBAAmB,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;YAEtG,MAAM,SAAS,GAAG;gBACjB,GAAG,EAAE,OAAO;gBACZ,UAAU;gBACV,OAAO,EAAE,IAAI;gBACb,MAAM,EAAE,KAAK;gBACb,MAAM,EAAE,mBAAmB;gBAC3B,SAAS;gBACT,cAAc,EAAE,QAAQ;gBACxB,OAAO,EAAE,4FAA4F;aACrG,CAAA;YAED,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,EAAE,cAAc,EAAE,uBAAuB,CAAC,EAAE,CAAC;gBAC9E,OAAO,EAAE,CAAA;YACV,CAAC;iBAAM,CAAC;gBACP,OAAO,EAAE,CAAA;YACV,CAAC;QACF,CAAC;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}

package/out/recipes/bare/postcode/eval.d.ts

@@ -5,7 +5,7 @@
  *
  *   Reserved bare-postcode capability cases. These strings are excluded from every country in the
  *   training recipe and are read only by evaluation. Reservation is by normalized input string,
- *   because `NNN NN` is shared by CZ, SK, SE and GR; excluding a Czech row alone could still train
+ *   because `NNN NN` is shared by CZ, SK, SE and GR. excluding only a Czech row could still train
  *   the model on the identical Swedish input.
  */
 export type BarePostcodeEvalFamily = "nnn_nn" | "nnnn_ll";

package/out/recipes/bare/postcode/eval.js

@@ -5,7 +5,7 @@
  *
  *   Reserved bare-postcode capability cases. These strings are excluded from every country in the
  *   training recipe and are read only by evaluation. Reservation is by normalized input string,
- *   because `NNN NN` is shared by CZ, SK, SE and GR; excluding a Czech row alone could still train
+ *   because `NNN NN` is shared by CZ, SK, SE and GR. excluding only a Czech row could still train
  *   the model on the identical Swedish input.
  */
 const SPACED_FIVE_CZ = [

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

@@ -0,0 +1,41 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ * `bare-postcode` — a postcode without another address component, in the country's format.
+ * Every surface the recipe renders is run through `detectKnownFormats` and refused unless the detector
+ * calls it a postcode, so the rendering table and `known-formats.ts`'s patterns must agree.
+ */
+import type { PathBuilderLike } from "path-ts";
+import { type CorpusRecipe } from "#recipes/scaffold";
+/**
+ * Choose distinct postcodes without inheriting the publisher's row order.
+ *
+ * A prior sort makes the result independent of input order.
+ * The seeded sample spreads reproducibly across the complete set.
+ */
+export declare function selectPostcodes(codes: Iterable<string>, limit: number, seed: number): string[];
+/**
+ * Return every required input path that is absent, preserving declaration order for diagnostics.
+ */
+export declare function findMissingPostcodeSources<P extends PathBuilderLike>(paths: readonly P[], exists?: (path: P) => Promise<boolean>): Promise<P[]>;
+/**
+ * Every surface a country writes for one postcode, spaced form first, or `[]`
+ * when this recipe defines no form for that country or the code does not fit.
+ *
+ * Exported so a test can pin that every rendered surface is one {@linkcode detectedAsPostcode} accepts.
+ */
+export declare function renderBarePostcode(country: string, postcode: string): string[];
+/**
+ * Whether `known-formats.ts` reads this surface as a postcode across its whole span.
+ *
+ * A surface the detector does not recognize would train the model on a string
+ * the query-shape prior cannot then support.
+ */
+export declare function detectedAsPostcode(surface: string): boolean;
+/**
+ * Recipe registered with the corpus builder.
+ */
+export declare const barePostcodeRecipe: CorpusRecipe;
+//# sourceMappingURL=postcode.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"postcode.d.ts","sourceRoot":"","sources":["../../../lib/recipes/bare/postcode.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAQH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAA;AAI9C,OAAO,EAAiB,KAAK,YAAY,EAAkC,MAAM,mBAAmB,CAAA;AA8GpG;;;;;GAKG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,QAAQ,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,CAM9F;AAED;;GAEG;AACH,wBAAsB,0BAA0B,CAAC,CAAC,SAAS,eAAe,EACzE,KAAK,EAAE,SAAS,CAAC,EAAE,EACnB,MAAM,GAAE,CAAC,IAAI,EAAE,CAAC,KAAK,OAAO,CAAC,OAAO,CAAc,GAChD,OAAO,CAAC,CAAC,EAAE,CAAC,CAId;AAED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,EAAE,CAM9E;AAED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAI3D;AAED;;GAEG;AACH,eAAO,MAAM,kBAAkB,EAAE,YAgKhC,CAAA"}

package/out/recipes/bare/{postcode/index.js → postcode.js}

@@ -3,41 +3,44 @@
  * @license AGPL-3.0
  * @author Teffen Ellis, et al.
  *
- *   `bare-postcode` — a postcode standing ALONE, in the form its country writes.
- *
- *   The shape is absent from the corpus. A census of the 685,187,151 rows the v0.29.0 mixture reads
- *   found ZERO two-token rows of `NNN NN` and one of `NNNN LL`, so `100 00` and `1012 LG` are out of
- *   distribution and every model's answer for them is generalization off longer lines. That
- *   generalization is what moved: held-out CZ/SK codes read 31/32 as a postcode under v5.0.0, v5.2.0
- *   AND v5.3.0 — a seed pair agreeing, so not training noise — and 0/32 under v5.4.0, whose added
- *   OpenStreetMap PK/BD/VN rows raise the leading-digit-is-a-house-number evidence at that opening
- *   from 393 to 444 samples per epoch.
- *
- *   Neither existing repair reaches it. `nl-postcode` and `cz-pcfirst-preposition` teach the same
- *   postcodes IN CONTEXT — every row they emit carries a street, a number and a city — and the
- *   12,759 in-context `NNN NN` rows do not transfer to the bare input. The deterministic half,
- *   `buildEmissionPriors`, caps near 0.95 at the default `biasScale` against a measured 1.67-to-3.78
- *   nat deficit.
- *
- *   So this recipe emits the postcode and NOTHING else, which is the one thing no sibling does.
- *
- *   VERIFIED AGAINST THE PRIOR, NOT MERELY MATCHED TO IT. Every surface is run through
- *   `detectKnownFormats` and refused unless the detector calls it a postcode. The recipe's rendering
- *   table and `known-formats.ts`'s patterns have to agree, and matched constants would not prove
- *   they do — the trained surface and the prior that boosts it come from one check.
+ * `bare-postcode` — a postcode without another address component, in the country's format.
+ * Every surface the recipe renders is run through `detectKnownFormats` and refused unless the detector
+ * calls it a postcode, so the rendering table and `known-formats.ts`'s patterns must agree.
  */
+import { isNLPostcodeKey } from "@mailwoman/codex/nl";
 import { dataRootPath } from "@mailwoman/core/data-root";
 import { pathExists } from "@mailwoman/core/fs/readers";
 import { SeededRandom } from "@mailwoman/core/utils";
 import { computeQueryShape } from "@mailwoman/query-shape";
 import { isPostcodeFormat } from "@mailwoman/query-shape/known-formats";
-import { join } from "path-ts";
+import { normalizeGauntletSurface, readGauntletInputs } from "#gauntlet-inputs";
 import { isReservedBarePostcode } from "#recipes/bare/postcode/eval";
-import { alignAndWrite, readCSVRecords, sliceSourceID } from "#recipes/scaffold";
+import { alignAndWrite, readCSVRecords, recipeSourceID } from "#recipes/scaffold";
+import { defaultRecipeSource } from "#recipes/sources";
+import { SourceRegister } from "#registers";
+import { SurfaceOrigin } from "#types";
+/**
+ * A postcode in the form its country writes, with no surrounding text.
+ *
+ * The codes come from this repository's postcode-format tables rather than a
+ * national postcode file, so the register is the codex.
+ */
+/**
+ * Resolved once, because `source_id` includes the source as its prefix and a pair
+ * that disagreed would name a source no row of this output declares.
+ */
+const BARE_POSTCODE_SOURCE = defaultRecipeSource("synth-bare-postcode");
+const BARE_POSTCODE_PROVENANCE = {
+    register: SourceRegister.Codex,
+    surface: SurfaceOrigin.Attested,
+    // The codex table is the record.
+    // Its rows have no `source_id` of their own.
+    baseSourceID: null,
+};
 const EXTRACTED = dataRootPath("openaddresses", "extracted");
 /**
- * Read from the archive rather than guessed: `readZippedCSVRecords` throws on a member that is not there, and
- * OpenAddresses keeps the Swedish spelling on two of these (`savsjö`, `Österåker`) while folding the rest to ASCII.
+ * Read from the archive rather than guessed, because OpenAddresses keeps the Swedish
+ * spelling on two of these (`savsjö`, `Österåker`) while folding the rest to ascii.
  */
 const SWEDISH_MUNICIPALITIES = [
     "alingsas",
@@ -58,51 +61,47 @@ const SWEDISH_MUNICIPALITIES = [
     "Österåker",
 ];
 /**
- * GREECE IS ABSENT ON PURPOSE. `gr/b/municipality_of_kalamaria.csv` is the archive's only Greek member and it declares
- * a `postcode` column carrying nothing: 0 values in 10,877 rows. `gr_postcode` shares `NNN NN` with the three below, so
- * a Greek reader is served by what they teach until a Greek source with postcodes exists — but no row here claims to be
- * Greek.
+ * Greece is deliberately absent: its only member declares a `postcode` column with no values over
+ * 10,877 rows, though `gr_postcode` shares `NNN NN` with CZ/SK/SE, so no row here claims to be Greek.
  */
 const SOURCES = [
-    { csv: join(EXTRACTED, "cz", "countrywide.csv"), country: "CZ" },
-    { csv: join(EXTRACTED, "sk", "countrywide.csv"), country: "SK" },
-    { csv: join(EXTRACTED, "nl", "countrywide.csv"), country: "NL" },
+    { csv: EXTRACTED("cz", "countrywide.csv"), country: "CZ" },
+    { csv: EXTRACTED("sk", "countrywide.csv"), country: "SK" },
+    { csv: EXTRACTED("nl", "countrywide.csv"), country: "NL" },
     ...SWEDISH_MUNICIPALITIES.map((name) => ({
-        csv: join(EXTRACTED, "se", `municipality_of_${name}.csv`),
+        csv: EXTRACTED("se", `municipality_of_${name}.csv`),
         country: "SE",
     })),
 ];
 /**
- * How a country writes its postcode when a person types it alone, and the locale to stamp.
+ * The postcode format to use on its own and the locale to stamp.
  *
- * `render` receives the source's own spelling with whitespace stripped, and answers every surface that country writes —
- * the spaced form first where one exists, because that is the failing one. A country whose written form is the source's
- * form answers a single entry.
- *
- * Only countries whose bare postcode COLLIDES with a house number are here: an all-digit or digits-then-letters opening
- * is what the model reads as `house_number`. `SW1A 1AA` opens with letters and was never in doubt, so GB is
- * deliberately absent.
+ * This map includes countries whose bare postcode collides with a house number.
+ * `render` answers with the spaced form first.
+ * GB is absent because `SW1A 1AA` starts with letters.
  */
 const WRITTEN_FORMS = new Map([
-    // `NNN NN`, written with the space. The four countries share the shape, which is why
-    // `detectKnownFormats` answers all four for one string and why the label is the same for each.
+    // `NNN NN`, written with the space.
+    // The four countries share the shape, so `detectKnownFormats` labels all four the same.
     ["CZ", { locale: "cs-CZ", render: spacedThree }],
     ["SK", { locale: "sk-SK", render: spacedThree }],
     ["SE", { locale: "sv-SE", render: spacedThree }],
     ["GR", { locale: "el-GR", render: spacedThree }],
-    // `NNNN LL`. Both spellings are attested and the spaced one is what fails.
+    // Both spellings of `nnnn LL` are attested.
+    // The spaced form is the failing case.
     [
         "NL",
         {
             locale: "nl-NL",
-            render: (compact) => /^\d{4}[A-Z]{2}$/.test(compact) ? [`${compact.slice(0, 4)} ${compact.slice(4)}`, compact] : [],
+            render: (compact) => (isNLPostcodeKey(compact) ? [`${compact.slice(0, 4)} ${compact.slice(4)}`, compact] : []),
         },
     ],
 ]);
 /**
- * `10000` → `["100 00", "10000"]`. Sweden and Greece write five digits with the space after the third, exactly as
- * Czechia and Slovakia do; the compact form rides along because sources store it that way and a reader types it both
- * ways.
+ * `10000` → `["100 00", "10000"]`.
+ *
+ * Sweden and Greece group five digits after the third, as Czechia and Slovakia do.
+ * The compact form remains because sources store it that way.
  */
 function spacedThree(compact) {
     if (!/^\d{5}$/.test(compact))
@@ -112,8 +111,8 @@ function spacedThree(compact) {
 /**
  * Choose distinct postcodes without inheriting the publisher's row order.
  *
- * Sorting first makes the result independent of input-file order; the seeded sample then gives a reproducible spread
- * across the complete set instead of taking the first municipality or numeric prefix that happens to fill the cap.
+ * A prior sort makes the result independent of input order.
+ * The seeded sample spreads reproducibly across the complete set.
  */
 export function selectPostcodes(codes, limit, seed) {
     const pool = [...new Set(codes)].toSorted();
@@ -129,12 +128,10 @@ export async function findMissingPostcodeSources(paths, exists = pathExists) {
     return results.filter((result) => !result.exists).map(({ path }) => path);
 }
 /**
- * Every surface a country writes for one postcode, spaced form first, or `[]` when this recipe carries no form for that
- * country or the code does not fit the one it carries.
+ * Every surface a country writes for one postcode, spaced form first, or `[]`
+ * when this recipe defines no form for that country or the code does not fit.
  *
- * Exported because it is the half of the recipe a test can reach: `run` reads a 500 MB archive from the data root, and
- * the contract worth pinning — that every surface this renders is one {@linkcode detectedAsPostcode} accepts — needs
- * neither.
+ * Exported so a test can pin that every rendered surface is one {@linkcode detectedAsPostcode} accepts.
  */
 export function renderBarePostcode(country, postcode) {
     const form = WRITTEN_FORMS.get(country.trim().toUpperCase());
@@ -145,19 +142,18 @@ export function renderBarePostcode(country, postcode) {
 /**
  * Whether `known-formats.ts` reads this surface as a postcode across its whole span.
  *
- * A surface the detector does not recognize would train the model on a string the query-shape prior cannot then
- * support, which is the divergence this recipe exists to close rather than widen.
+ * A surface the detector does not recognize would train the model on a string
+ * the query-shape prior cannot then support.
  */
 export function detectedAsPostcode(surface) {
     return computeQueryShape(surface).knownFormats.some((hit) => isPostcodeFormat(hit.format) && hit.span.start === 0 && hit.span.end === surface.length);
 }
 /**
- * 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.
+ * Recipe registered with the corpus builder.
  */
 export const barePostcodeRecipe = {
     name: "bare-postcode",
-    description: "A postcode alone, in its country's written form — the shape no slice carries (CZ/SK/SE/NL)",
+    description: "A postcode alone, in its country's written form — the shape no other recipe output carries (CZ/SK/SE/NL)",
     mode: "generate",
     options: [
         {
@@ -170,22 +166,26 @@ export const barePostcodeRecipe = {
         let emitted = 0;
         let skipped = 0;
         let unrecognized = 0;
-        // Check the complete input set before writing the first row. A missing municipality otherwise
-        // produces a plausible non-empty artifact with less Swedish coverage than the recipe declares.
-        const missing = await findMissingPostcodeSources(SOURCES.map(({ csv }) => String(csv)));
+        // Preflight the complete input set before writing a row.
+        // A missing municipality would otherwise produce a plausible artifact with
+        // less Swedish coverage than the recipe declares.
+        const missing = await findMissingPostcodeSources(SOURCES.map(({ csv }) => csv));
         if (missing.length) {
             throw new Error(`bare-postcode is missing ${missing.length} of ${SOURCES.length} required OpenAddresses files, ` +
                 `starting with ${missing[0]}. Fetch the per-source runs into \`openaddresses/extracted/\` ` +
                 "(anonymous, see packages/corpus/CLAUDE.md).");
         }
-        // A PER-COUNTRY budget, because supply is wildly uneven and the shortage is where the capability
-        // broke: the Netherlands publishes ~460,000 distinct `NNNN LL` codes against Czechia's 2,669 and
-        // Slovakia's 1,059, so an uncapped pass emits 98.9% Dutch rows and teaches the `NNN NN` countries —
-        // the ones reading 0/32 — almost nothing. Equal shares, each country keeping whatever it can fill.
+        // Per-country budget, because supply is uneven: the Netherlands publishes ~460,000
+        // distinct `nnnn LL` codes against Czechia's 2,669 and Slovakia's 1,059, so an uncapped
+        // pass would emit 98.9% Dutch rows and teach the `NNN NN` countries almost no pattern.
         const countries = [...new Set(SOURCES.map((source) => source.country))];
         const budget = opts.count ? Math.ceil(opts.count / countries.length) : Number.POSITIVE_INFINITY;
         const perCountry = new Map(countries.map((country) => [country, 0]));
         const codesByCountry = new Map(countries.map((country) => [country, new Set()]));
+        // The gauntlet boards are a second held-out register, separate from
+        // `BARE_POSTCODE_EVAL_CASES`, and `cz/bare-postcode.jsonl` predates this recipe.
+        const boardInputs = await readGauntletInputs();
+        let boardRefused = 0;
         for (const source of SOURCES) {
             const form = WRITTEN_FORMS.get(source.country);
             if (!form)
@@ -201,6 +201,14 @@ export const barePostcodeRecipe = {
                     skipped++;
                     continue;
                 }
+                // A board row is refused on any of its written forms, so the check runs over
+                // each rendered surface rather than the compact code; `normalizeGauntletSurface`
+                // makes `100 00` and `10000` one string.
+                if (form.render(compact).some((surface) => boardInputs.has(normalizeGauntletSurface(surface)))) {
+                    boardRefused++;
+                    skipped++;
+                    continue;
+                }
                 if (!form.render(compact).length) {
                     skipped++;
                     continue;
@@ -218,8 +226,8 @@ export const barePostcodeRecipe = {
                 for (const surface of surfaces) {
                     if (perCountry.get(country) >= budget)
                         break;
-                    // A surface the detector does not read as a postcode is a disagreement between this table
-                    // and `known-formats.ts`, and emitting it would teach a string the prior cannot support.
+                    // A surface the detector does not read as a postcode is a disagreement
+                    // between this table and `known-formats.ts`.
                     if (!detectedAsPostcode(surface)) {
                         unrecognized++;
                         continue;
@@ -230,15 +238,15 @@ export const barePostcodeRecipe = {
                         components,
                         country,
                         locale: form.locale,
-                        source: "synth-bare-postcode",
-                        source_id: sliceSourceID("synth-bare-postcode", {
+                        source: BARE_POSTCODE_SOURCE,
+                        source_id: recipeSourceID(BARE_POSTCODE_SOURCE, {
                             ...components,
                             c: country,
                         }),
                         corpus_version: "0.30.0",
                         license: "Synthetic — bare-postcode; postcodes from OpenAddresses (per-source attribution in the model card)",
                     };
-                    if (alignAndWrite(write, canonical, "bare-postcode")) {
+                    if (alignAndWrite(write, canonical, "bare-postcode", BARE_POSTCODE_PROVENANCE)) {
                         emitted++;
                         perCountry.set(country, perCountry.get(country) + 1);
                     }
@@ -248,20 +256,24 @@ export const barePostcodeRecipe = {
                 }
             }
         }
-        // A surface this table renders and the detector refuses is a contract break, not a data quirk: every
-        // form here is one `known-formats.ts` declares a pattern for, so the count is expected to be zero and
-        // a non-zero one names which side moved.
+        // A rendered surface the detector refuses is an interface failure: every form here is
+        // one `known-formats.ts` declares a pattern for, so the count is expected to be zero.
         if (unrecognized > 0) {
             throw new Error(`${unrecognized} rendered surfaces were not read as a postcode by detectKnownFormats — the ` +
                 "WRITTEN_FORMS table and known-formats.ts's PATTERNS disagree. Reconcile them before building.");
         }
-        // AN EMPTY BUILD IS A FAILURE, NOT AN EMPTY ANSWER. Required files were preflighted above, so zero
-        // rows here means their postcode columns or the written-form rules no longer provide usable data.
+        // An empty build is a failure rather than an empty answer: the required
+        // files were preflighted above, so zero rows means their postcode columns
+        // or the written-form rules stopped providing usable data.
         if (emitted === 0) {
             throw new Error(`bare-postcode emitted no rows from ${SOURCES.length} readable sources. ` +
                 "The postcode columns or the written-form rules changed.");
         }
+        // Reported rather than folded into `skipped` so a growing board is visible.
+        // A count of zero where the boards hold postcodes means the check stopped reaching them.
+        console.error(`  bare-postcode: ${boardRefused.toLocaleString()} rows refused as gauntlet board inputs ` +
+            `(${boardInputs.size.toLocaleString()} inputs read)`);
         return { read, emitted, skipped };
     },
 };
-//# sourceMappingURL=index.js.map
+//# sourceMappingURL=postcode.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"postcode.js","sourceRoot":"","sources":["../../../lib/recipes/bare/postcode.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAA;AACrD,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAA;AACxD,OAAO,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAA;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAA;AACpD,OAAO,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAA;AAC1D,OAAO,EAAE,gBAAgB,EAAE,MAAM,sCAAsC,CAAA;AAGvE,OAAO,EAAE,wBAAwB,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAA;AAC/E,OAAO,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAA;AACpE,OAAO,EAAE,aAAa,EAAqB,cAAc,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAA;AACpG,OAAO,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAA;AACtD,OAAO,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAC3C,OAAO,EAAE,aAAa,EAAE,MAAM,QAAQ,CAAA;AAEtC;;;;;GAKG;AACH;;;GAGG;AACH,MAAM,oBAAoB,GAAG,mBAAmB,CAAC,qBAAqB,CAAC,CAAA;AAEvE,MAAM,wBAAwB,GAAG;IAChC,QAAQ,EAAE,cAAc,CAAC,KAAK;IAC9B,OAAO,EAAE,aAAa,CAAC,QAAQ;IAC/B,iCAAiC;IACjC,6CAA6C;IAC7C,YAAY,EAAE,IAAI;CAClB,CAAA;AAWD,MAAM,SAAS,GAAG,YAAY,CAAC,eAAe,EAAE,WAAW,CAAC,CAAA;AAE5D;;;GAGG;AACH,MAAM,sBAAsB,GAAG;IAC9B,UAAU;IACV,UAAU;IACV,UAAU;IACV,aAAa;IACb,SAAS;IACT,QAAQ;IACR,cAAc;IACd,OAAO;IACP,OAAO;IACP,QAAQ;IACR,WAAW;IACX,SAAS;IACT,UAAU;IACV,SAAS;IACT,OAAO;IACP,WAAW;CACX,CAAA;AAED;;;GAGG;AACH,MAAM,OAAO,GAAqB;IACjC,EAAE,GAAG,EAAE,SAAS,CAAC,IAAI,EAAE,iBAAiB,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE;IAC1D,EAAE,GAAG,EAAE,SAAS,CAAC,IAAI,EAAE,iBAAiB,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE;IAC1D,EAAE,GAAG,EAAE,SAAS,CAAC,IAAI,EAAE,iBAAiB,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE;IAC1D,GAAG,sBAAsB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QACxC,GAAG,EAAE,SAAS,CAAC,IAAI,EAAE,mBAAmB,IAAI,MAAM,CAAC;QACnD,OAAO,EAAE,IAAI;KACb,CAAC,CAAC;CACH,CAAA;AAED;;;;;;GAMG;AACH,MAAM,aAAa,GAAmF,IAAI,GAAG,CAAC;IAC7G,oCAAoC;IACpC,wFAAwF;IACxF,CAAC,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;IAChD,CAAC,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;IAChD,CAAC,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;IAChD,CAAC,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;IAChD,4CAA4C;IAC5C,uCAAuC;IACvC;QACC,IAAI;QACJ;YACC,MAAM,EAAE,OAAO;YACf,MAAM,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;SAC9G;KACD;CACD,CAAC,CAAA;AAEF;;;;;GAKG;AACH,SAAS,WAAW,CAAC,OAAe;IACnC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,OAAO,EAAE,CAAA;IAEvC,OAAO,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,OAAO,CAAC,CAAA;AAC/D,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAAC,KAAuB,EAAE,KAAa,EAAE,IAAY;IACnF,MAAM,IAAI,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAA;IAE3C,IAAI,KAAK,IAAI,IAAI,CAAC,MAAM;QAAE,OAAO,IAAI,CAAA;IAErC,OAAO,IAAI,YAAY,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAA;AAC/D,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAC/C,KAAmB,EACnB,MAAM,GAAkC,UAAU;IAElD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAA;IAEpG,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC,IAAI,CAAC,CAAA;AAC1E,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,kBAAkB,CAAC,OAAe,EAAE,QAAgB;IACnE,MAAM,IAAI,GAAG,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAA;IAE5D,IAAI,CAAC,IAAI;QAAE,OAAO,EAAE,CAAA;IAEpB,OAAO,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAA;AAC1E,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,kBAAkB,CAAC,OAAe;IACjD,OAAO,iBAAiB,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,IAAI,CAClD,CAAC,GAAG,EAAE,EAAE,CAAC,gBAAgB,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,KAAK,OAAO,CAAC,MAAM,CAChG,CAAA;AACF,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAiB;IAC/C,IAAI,EAAE,eAAe;IACrB,WAAW,EACV,0GAA0G;IAC3G,IAAI,EAAE,UAAU;IAChB,OAAO,EAAE;QACR;YACC,IAAI,EAAE,SAAS;YACf,WAAW,EAAE,sFAAsF;SACnG;KACD;IACD,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK;QACpB,IAAI,IAAI,GAAG,CAAC,CAAA;QACZ,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,OAAO,GAAG,CAAC,CAAA;QACf,IAAI,YAAY,GAAG,CAAC,CAAA;QAEpB,yDAAyD;QACzD,2EAA2E;QAC3E,kDAAkD;QAClD,MAAM,OAAO,GAAG,MAAM,0BAA0B,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,EAAE,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAA;QAE/E,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;YACpB,MAAM,IAAI,KAAK,CACd,4BAA4B,OAAO,CAAC,MAAM,OAAO,OAAO,CAAC,MAAM,iCAAiC;gBAC/F,iBAAiB,OAAO,CAAC,CAAC,CAAC,gEAAgE;gBAC3F,6CAA6C,CAC9C,CAAA;QACF,CAAC;QAED,mFAAmF;QACnF,wFAAwF;QACxF,uFAAuF;QACvF,MAAM,SAAS,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAA;QACvE,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,iBAAiB,CAAA;QAC/F,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC,CAAA;QACpE,MAAM,cAAc,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,IAAI,GAAG,EAAU,CAAC,CAAC,CAAC,CAAA;QAExF,oEAAoE;QACpE,iFAAiF;QACjF,MAAM,WAAW,GAAG,MAAM,kBAAkB,EAAE,CAAA;QAC9C,IAAI,YAAY,GAAG,CAAC,CAAA;QAEpB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;YAC9B,MAAM,IAAI,GAAG,aAAa,CAAC,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;YAE9C,IAAI,CAAC,IAAI;gBAAE,SAAQ;YAEnB,IAAI,KAAK,EAAE,MAAM,GAAG,IAAI,cAAc,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC;gBACpD,IAAI,EAAE,CAAA;gBAEN,MAAM,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC;qBACxC,IAAI,EAAE;qBACN,WAAW,EAAE;qBACb,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAA;gBAEzB,MAAM,YAAY,GAAG,cAAc,CAAC,GAAG,CAAC,MAAM,CAAC,OAAO,CAAE,CAAA;gBAExD,IAAI,CAAC,OAAO,IAAI,sBAAsB,CAAC,OAAO,CAAC,IAAI,YAAY,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;oBAC9E,OAAO,EAAE,CAAA;oBAET,SAAQ;gBACT,CAAC;gBAED,6EAA6E;gBAC7E,iFAAiF;gBACjF,yCAAyC;gBACzC,IAAI,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,WAAW,CAAC,GAAG,CAAC,wBAAwB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;oBAChG,YAAY,EAAE,CAAA;oBAEd,OAAO,EAAE,CAAA;oBAET,SAAQ;gBACT,CAAC;gBAED,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,CAAC;oBAClC,OAAO,EAAE,CAAA;oBAET,SAAQ;gBACT,CAAC;gBAED,YAAY,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;YAC1B,CAAC;QACF,CAAC;QAED,KAAK,MAAM,OAAO,IAAI,SAAS,EAAE,CAAC;YACjC,MAAM,IAAI,GAAG,aAAa,CAAC,GAAG,CAAC,OAAO,CAAE,CAAA;YACxC,MAAM,YAAY,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,iBAAiB,CAAA;YAC/F,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,GAAG,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC,CAAA;YACpF,MAAM,QAAQ,GAAG,eAAe,CAAC,cAAc,CAAC,GAAG,CAAC,OAAO,CAAE,EAAE,YAAY,EAAE,WAAW,CAAC,CAAA;YAEzF,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;gBAChC,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;gBAErC,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;oBAChC,IAAI,UAAU,CAAC,GAAG,CAAC,OAAO,CAAE,IAAI,MAAM;wBAAE,MAAK;oBAE7C,uEAAuE;oBACvE,6CAA6C;oBAC7C,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,EAAE,CAAC;wBAClC,YAAY,EAAE,CAAA;wBAEd,SAAQ;oBACT,CAAC;oBAED,MAAM,UAAU,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAA;oBAExC,MAAM,SAAS,GAAG;wBACjB,GAAG,EAAE,OAAO;wBACZ,UAAU;wBACV,OAAO;wBACP,MAAM,EAAE,IAAI,CAAC,MAAM;wBACnB,MAAM,EAAE,oBAAoB;wBAC5B,SAAS,EAAE,cAAc,CAAC,oBAAoB,EAAE;4BAC/C,GAAG,UAAU;4BACb,CAAC,EAAE,OAAO;yBACV,CAAC;wBACF,cAAc,EAAE,QAAQ;wBACxB,OAAO,EACN,oGAAoG;qBACrG,CAAA;oBAED,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,wBAAwB,CAAC,EAAE,CAAC;wBAChF,OAAO,EAAE,CAAA;wBACT,UAAU,CAAC,GAAG,CAAC,OAAO,EAAE,UAAU,CAAC,GAAG,CAAC,OAAO,CAAE,GAAG,CAAC,CAAC,CAAA;oBACtD,CAAC;yBAAM,CAAC;wBACP,OAAO,EAAE,CAAA;oBACV,CAAC;gBACF,CAAC;YACF,CAAC;QACF,CAAC;QAED,sFAAsF;QACtF,sFAAsF;QACtF,IAAI,YAAY,GAAG,CAAC,EAAE,CAAC;YACtB,MAAM,IAAI,KAAK,CACd,GAAG,YAAY,6EAA6E;gBAC3F,+FAA+F,CAChG,CAAA;QACF,CAAC;QAED,wEAAwE;QACxE,0EAA0E;QAC1E,2DAA2D;QAC3D,IAAI,OAAO,KAAK,CAAC,EAAE,CAAC;YACnB,MAAM,IAAI,KAAK,CACd,sCAAsC,OAAO,CAAC,MAAM,qBAAqB;gBACxE,yDAAyD,CAC1D,CAAA;QACF,CAAC;QAED,4EAA4E;QAC5E,yFAAyF;QACzF,OAAO,CAAC,KAAK,CACZ,oBAAoB,YAAY,CAAC,cAAc,EAAE,yCAAyC;YACzF,IAAI,WAAW,CAAC,IAAI,CAAC,cAAc,EAAE,eAAe,CACrD,CAAA;QAED,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClC,CAAC;CACD,CAAA"}

package/out/recipes/boundary-stress.d.ts

@@ -3,18 +3,18 @@
  * @license AGPL-3.0
  * @author Teffen Ellis, et al.
  *
- *   `boundary-stress` slice recipe (#375) — the boundary-instability augmentation. Self-generates
- *   `--count` rows from {@link synthesizeBoundaryStressRow}'s weighted template mix (the v1.7.0,
- *   DeepSeek-tuned 2026-06-18 composition), aligns each to BIO, and emits a labeled JSONL. The
- *   change for the taxonomy's #1 parser family (the boundary-wobble class). Ported from
- *   scripts/build-boundary-stress-slice.mjs.
+ *   `boundary-stress` recipe, the boundary-instability augmentation. Self-generates `--count` rows
+ *   from {@link synthesizeBoundaryStressRow}'s weighted template mix. It aligns each to BIO and emits a
+ *   labeled jsonl.
  *
- *   `synthesizeBoundaryStressRow` is NOT re-exported from the corpus index — imported directly here.
+ *   `synthesizeBoundaryStressRow` is imported directly here.
  */
 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.
+ * Recipe registered with the corpus builder.
+ *
+ * See the file header for the parse behavior this recipe exercises and `description`
+ * below for the surface form it generates.
  */
 export declare const boundaryStressRecipe: CorpusRecipe;
 //# sourceMappingURL=boundary-stress.d.ts.map

package/out/recipes/boundary-stress.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"boundary-stress.d.ts","sourceRoot":"","sources":["../../lib/recipes/boundary-stress.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAIH,OAAO,EAAE,KAAK,YAAY,EAAiB,MAAM,mBAAmB,CAAA;AAoCpE;;;GAGG;AACH,eAAO,MAAM,oBAAoB,EAAE,YA6ClC,CAAA"}
+{"version":3,"file":"boundary-stress.d.ts","sourceRoot":"","sources":["../../lib/recipes/boundary-stress.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAKH,OAAO,EAAE,KAAK,YAAY,EAAkB,MAAM,mBAAmB,CAAA;AA2CrE;;;;;GAKG;AACH,eAAO,MAAM,oBAAoB,EAAE,YAsDlC,CAAA"}

package/out/recipes/boundary-stress.js

@@ -3,24 +3,30 @@
  * @license AGPL-3.0
  * @author Teffen Ellis, et al.
  *
- *   `boundary-stress` slice recipe (#375) — the boundary-instability augmentation. Self-generates
- *   `--count` rows from {@link synthesizeBoundaryStressRow}'s weighted template mix (the v1.7.0,
- *   DeepSeek-tuned 2026-06-18 composition), aligns each to BIO, and emits a labeled JSONL. The
- *   change for the taxonomy's #1 parser family (the boundary-wobble class). Ported from
- *   scripts/build-boundary-stress-slice.mjs.
+ *   `boundary-stress` recipe, the boundary-instability augmentation. Self-generates `--count` rows
+ *   from {@link synthesizeBoundaryStressRow}'s weighted template mix. It aligns each to BIO and emits a
+ *   labeled jsonl.
  *
- *   `synthesizeBoundaryStressRow` is NOT re-exported from the corpus index — imported directly here.
+ *   `synthesizeBoundaryStressRow` is imported directly here.
  */
+import { stringifyJSON } from "@mailwoman/core/json";
 import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils";
-import { sliceSourceID } from "#recipes/scaffold";
+import { recipeSourceID } from "#recipes/scaffold";
+import { defaultRecipeSource } from "#recipes/sources";
 import { synthesizeBoundaryStressRow } from "#synthesizers/boundary-stress";
+import { SurfaceOrigin } from "#types";
 import { alignRow } from "#utils";
 /**
- * Revised composition (v1.7.0, DeepSeek-tuned 2026-06-18): `bare-locality` ~11% (recover the 84% locality drop on bare
- * "City, STATE" rows WITHOUT becoming a locality-first majority), and house-number-before:after = 7:3 (FR's dominant
- * order is number-BEFORE; 30% after breaks the order-bias shortcut without risking FR hn-before accuracy). The three
- * original non-number shapes keep the bulk. Weights sum to 1.0. Key order is required — it drives the cumulative
- * thresholds below.
+ * Template weights, in the order the cumulative thresholds below read.
+ *
+ * The mix keeps `bare-locality` near 11%, so bare "City, state" rows are well
+ * represented without a locality-first majority.
+ * House-number-before and house-number-after render at 7:3.
+ *
+ * This breaks the order-bias shortcut while keeping FR house-number-before accuracy.
+ *
+ * Weights sum to 1.0.
+ * The key order drives the cumulative thresholds below.
  */
 const WEIGHTS = {
     "street-eats-affix": 0.22,
@@ -42,15 +48,16 @@ function pickTemplate(r) {
     return CUM.at(-1)[0];
 }
 /**
- * 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.
+ * Recipe registered with the corpus builder.
+ *
+ * See the file header for the parse behavior this recipe exercises and `description`
+ * below for the surface form it generates.
  */
 export const boundaryStressRecipe = {
     name: "boundary-stress",
     description: "Boundary-instability rows (#375): weighted template mix → synthesizeBoundaryStressRow → aligned BIO",
     mode: "generate",
     async run(opts, write) {
-        // Emit PRNG: the legacy build-boundary-stress-slice.mjs seeded mulberry32(opts.seed).
         const random = makeMulberry32(opts.seed);
         const count = opts.count ?? 20_000;
         let emitted = 0;
@@ -58,13 +65,16 @@ export const boundaryStressRecipe = {
         for (let i = 0; i < count; i++) {
             const row = synthesizeBoundaryStressRow(undefined, { random, forceTemplate: pickTemplate(random) });
             const country = row.locale.split("-")[1] ?? "US";
-            const source_id = sliceSourceID("synth-boundary-stress", { ...row.components, v: String(i) });
+            // One resolution for both, because `source_id` includes the source as its prefix
+            // and a pair that disagreed would name a source no row of this output declares.
+            const source = defaultRecipeSource("synth-boundary-stress");
+            const source_id = recipeSourceID(source, { ...row.components, v: String(i) });
             const canonical = {
                 raw: row.raw,
                 components: row.components,
                 country,
                 locale: row.locale,
-                source: "synth-boundary-stress",
+                source,
                 source_id,
                 corpus_version: "0.6.0",
                 license: "Synthetic — boundary-stress; derived from public-domain locality/region tuples",
@@ -74,8 +84,15 @@ export const boundaryStressRecipe = {
                 skipped++;
                 continue;
             }
-            // Match the base corpus parquet schema: flat synth_method / synth_base_id, not a nested `synth`.
-            write(JSON.stringify({ ...aligned.row, synth_method: `boundary-stress:${row.template}`, synth_base_id: null }) + "\n");
+            // Match the base corpus parquet schema: flat columns rather than a nested object.
+            write(stringifyJSON({
+                ...aligned.row,
+                recipe: `boundary-stress:${row.template}`,
+                base_source_id: null,
+                // Every component is drawn from a template mix built to sit on a tag boundary.
+                register: null,
+                surface: SurfaceOrigin.Invented,
+            }));
             emitted++;
         }
         return { emitted, skipped };