@mailwoman/codex 9.3.0 → 10.0.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mailwoman/codex",
-  "version": "9.3.0",
+  "version": "10.0.0",
   "description": "Per-address-system postal reference data + branded types (USPS street suffixes, US ZIP codes). Pure, zero-runtime-dep — the shared canonical home for postal-system primitives the parser, resolver, and synthesis layers all reach for.",
   "license": "AGPL-3.0-only OR LicenseRef-Commercial",
   "repository": {
@@ -23,6 +23,8 @@
     "!**/*.test.tsx",
     "lib/us/street-suffix.json",
     "out/us/street-suffix.json",
+    "lib/postcode/shapes.json",
+    "out/postcode/shapes.json",
     "!test/**"
   ],
   "type": "module",
@@ -85,10 +87,45 @@
       "node": "./out/nz/index.js",
       "default": "./out/nz/index.js"
     },
-    "./tools": {
-      "types": "./out/tools/index.d.ts",
-      "node": "./out/tools/index.js",
-      "default": "./out/tools/index.js"
+    "./address-layout": {
+      "types": "./out/address/layout.d.ts",
+      "node": "./out/address/layout.js",
+      "default": "./out/address/layout.js"
+    },
+    "./address-layouts": {
+      "types": "./out/address/layouts/index.d.ts",
+      "node": "./out/address/layouts/index.js",
+      "default": "./out/address/layouts/index.js"
+    },
+    "./address-layouts-generated": {
+      "types": "./out/address/layouts/generated.d.ts",
+      "node": "./out/address/layouts/generated.js",
+      "default": "./out/address/layouts/generated.js"
+    },
+    "./address-render": {
+      "types": "./out/address/render.d.ts",
+      "node": "./out/address/render.js",
+      "default": "./out/address/render.js"
+    },
+    "./address-format": {
+      "types": "./out/address/format.d.ts",
+      "node": "./out/address/format.js",
+      "default": "./out/address/format.js"
+    },
+    "./address-key": {
+      "types": "./out/address/key.d.ts",
+      "node": "./out/address/key.js",
+      "default": "./out/address/key.js"
+    },
+    "./component": {
+      "types": "./out/component.d.ts",
+      "node": "./out/component.js",
+      "default": "./out/component.js"
+    },
+    "./placetype-map": {
+      "types": "./out/placetype-map.d.ts",
+      "node": "./out/placetype-map.js",
+      "default": "./out/placetype-map.js"
     },
     "./normalize": {
       "types": "./out/normalize.d.ts",
@@ -106,19 +143,24 @@
       "default": "./out/it/index.js"
     },
     "./address-system-conventions": {
-      "types": "./out/address-system-conventions.d.ts",
-      "node": "./out/address-system-conventions.js",
-      "default": "./out/address-system-conventions.js"
+      "types": "./out/address/system-conventions.d.ts",
+      "node": "./out/address/system-conventions.js",
+      "default": "./out/address/system-conventions.js"
     },
     "./level-semantics": {
       "types": "./out/level-semantics.d.ts",
       "node": "./out/level-semantics.js",
       "default": "./out/level-semantics.js"
     },
+    "./postcode-shapes": {
+      "types": "./out/postcode/shapes.d.ts",
+      "node": "./out/postcode/shapes.js",
+      "default": "./out/postcode/shapes.js"
+    },
     "./postcode-systems": {
-      "types": "./out/postcode-systems.d.ts",
-      "node": "./out/postcode-systems.js",
-      "default": "./out/postcode-systems.js"
+      "types": "./out/postcode/systems.d.ts",
+      "node": "./out/postcode/systems.js",
+      "default": "./out/postcode/systems.js"
     },
     "./us/floor-designator": {
       "types": "./out/us/floor-designator.d.ts",
@@ -131,14 +173,19 @@
       "default": "./out/us/state.js"
     },
     "./us/street-suffix": {
-      "types": "./out/us/street-suffix.d.ts",
-      "node": "./out/us/street-suffix.js",
-      "default": "./out/us/street-suffix.js"
+      "types": "./out/us/street/suffix.d.ts",
+      "node": "./out/us/street/suffix.js",
+      "default": "./out/us/street/suffix.js"
     },
     "./us/unit-designator": {
       "types": "./out/us/unit-designator.d.ts",
       "node": "./out/us/unit-designator.js",
       "default": "./out/us/unit-designator.js"
+    },
+    "./kr": {
+      "types": "./out/kr/index.d.ts",
+      "node": "./out/kr/index.js",
+      "default": "./out/kr/index.js"
     }
   },
   "publishConfig": {
@@ -195,10 +242,45 @@
         "node": "./out/nz/index.js",
         "default": "./out/nz/index.js"
       },
-      "./tools": {
-        "types": "./out/tools/index.d.ts",
-        "node": "./out/tools/index.js",
-        "default": "./out/tools/index.js"
+      "./address-layout": {
+        "types": "./out/address/layout.d.ts",
+        "node": "./out/address/layout.js",
+        "default": "./out/address/layout.js"
+      },
+      "./address-layouts": {
+        "types": "./out/address/layouts/index.d.ts",
+        "node": "./out/address/layouts/index.js",
+        "default": "./out/address/layouts/index.js"
+      },
+      "./address-layouts-generated": {
+        "types": "./out/address/layouts/generated.d.ts",
+        "node": "./out/address/layouts/generated.js",
+        "default": "./out/address/layouts/generated.js"
+      },
+      "./address-render": {
+        "types": "./out/address/render.d.ts",
+        "node": "./out/address/render.js",
+        "default": "./out/address/render.js"
+      },
+      "./address-format": {
+        "types": "./out/address/format.d.ts",
+        "node": "./out/address/format.js",
+        "default": "./out/address/format.js"
+      },
+      "./address-key": {
+        "types": "./out/address/key.d.ts",
+        "node": "./out/address/key.js",
+        "default": "./out/address/key.js"
+      },
+      "./component": {
+        "types": "./out/component.d.ts",
+        "node": "./out/component.js",
+        "default": "./out/component.js"
+      },
+      "./placetype-map": {
+        "types": "./out/placetype-map.d.ts",
+        "node": "./out/placetype-map.js",
+        "default": "./out/placetype-map.js"
       },
       "./normalize": {
         "types": "./out/normalize.d.ts",
@@ -216,19 +298,24 @@
         "default": "./out/it/index.js"
       },
       "./address-system-conventions": {
-        "types": "./out/address-system-conventions.d.ts",
-        "node": "./out/address-system-conventions.js",
-        "default": "./out/address-system-conventions.js"
+        "types": "./out/address/system-conventions.d.ts",
+        "node": "./out/address/system-conventions.js",
+        "default": "./out/address/system-conventions.js"
       },
       "./level-semantics": {
         "types": "./out/level-semantics.d.ts",
         "node": "./out/level-semantics.js",
         "default": "./out/level-semantics.js"
       },
+      "./postcode-shapes": {
+        "types": "./out/postcode/shapes.d.ts",
+        "node": "./out/postcode/shapes.js",
+        "default": "./out/postcode/shapes.js"
+      },
       "./postcode-systems": {
-        "types": "./out/postcode-systems.d.ts",
-        "node": "./out/postcode-systems.js",
-        "default": "./out/postcode-systems.js"
+        "types": "./out/postcode/systems.d.ts",
+        "node": "./out/postcode/systems.js",
+        "default": "./out/postcode/systems.js"
       },
       "./us/floor-designator": {
         "types": "./out/us/floor-designator.d.ts",
@@ -241,14 +328,19 @@
         "default": "./out/us/state.js"
       },
       "./us/street-suffix": {
-        "types": "./out/us/street-suffix.d.ts",
-        "node": "./out/us/street-suffix.js",
-        "default": "./out/us/street-suffix.js"
+        "types": "./out/us/street/suffix.d.ts",
+        "node": "./out/us/street/suffix.js",
+        "default": "./out/us/street/suffix.js"
       },
       "./us/unit-designator": {
         "types": "./out/us/unit-designator.d.ts",
         "node": "./out/us/unit-designator.js",
         "default": "./out/us/unit-designator.js"
+      },
+      "./kr": {
+        "types": "./out/kr/index.d.ts",
+        "node": "./out/kr/index.js",
+        "default": "./out/kr/index.js"
       }
     },
     "imports": {
@@ -263,8 +355,6 @@
     "type-fest": "^5.9.0"
   },
   "devDependencies": {
-    "@mailwoman/annotations": "9.3.0",
-    "@mailwoman/core": "9.3.0",
-    "path-ts": "^2.3.0"
+    "@mailwoman/annotations": "10.0.0"
   }
 }

package/lib/tools/build-country-surface-lexicon.ts

@@ -1,177 +0,0 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   Build the COUNTRY-SURFACE LEXICON for the country-lexicon soft-feed channel (#1104). This is the
- *   third atlas channel, a sibling of the postcode anchor (#239/#240) and the gazetteer anchor
- *   (#464): a per-token multi-hot clue the neural GRAMMAR conditions on but never obeys. Country is a
- *   CLOSED, ENUMERABLE class (~250 surfaces) — atlas, not grammar — so a dictionary phrase-lookup
- *   recovers the WOF-admin / resolver hierarchy case ("United States of America, Wyoming, <locality>")
- *   the learned tagger reads as a leading STREET. Pelias handled the same class the same way
- *   (`WhosOnFirstClassifier extends PhraseClassifier`); this is the model-first analogue.
- *
- *   WHY A DEDICATED LEXICON (not just the gazetteer's `country` slot): the gazetteer already carries
- *   these surfaces in slot 0, and the shipped model already consumes them — yet the WOF-admin case
- *   still fails (model-card #1104: golden country recall 82.0% vs 88.6%). The country bit is one of a
- *   5-hot vector sharing ONE learned projection with region/po_box/cedex/homograph, and it is ZEROED
- *   adjacent to a postcode by `suppress_gazetteer_near_postcode` (exactly where a trailing "…12345
- *   USA" sits). A dedicated channel de-entangles the country signal (its own projection + confidence
- *   weight) and is immune to that suppression. See
- *   docs/superpowers/plans/2026-07-14-country-lexicon-channel.md.
- *
- *   The matcher REUSES the gazetteer's phrase-scan (longest-first n-gram over whitespace words,
- *   case-insensitive `entries` + uppercase-exact `code_entries`, char→piece projection) — one tested
- *   algorithm, two vocabularies. Only the vocabulary + the emitted feature differ. The emitted
- *   feature is 2-dim per piece: `[country_surface, country_ambiguous]`.
- *
- *   - `country_surface` (bit 1): the piece is part of a recognized country surface phrase.
- *   - `country_ambiguous` (bit 2): the SURFACE is a homograph (also a US region) or a common-word
- *     name ("Georgia", "America", "England", "IN") — a SOFT version of Pelias's hard blacklist. The
- *     model learns to trust `surface & !ambiguous` (unambiguous long/code forms) strongly and
- *     `surface & ambiguous` weakly, using context — model-first, never a hard drop, so recall on
- *     "Republic of Georgia" is preserved.
- *
- *   Source of truth: `@mailwoman/codex` (COUNTRY_SURFACE_FORMS + ISO2_TO_NAME) — the SAME data the
- *   corpus-python bridge `country-surfaces.json` is generated from (export-country-surfaces.ts), so
- *   the channel and the corpus extract synthesizer cannot diverge on what a country surface IS.
- *
- *   Output: data/gazetteer/country-surface-lexicon-v1.json (small, committed, provenance-tracked).
- *   Regenerate: `node codex/tools/build-country-surface-lexicon.ts`
- */
-
-import { makeDirectories, writeLocalTextFile } from "@mailwoman/core/fs/writers"
-import { repoRootPath } from "@mailwoman/core/paths"
-import { dirname } from "path-ts"
-
-import { COUNTRY_SURFACE_FORMS, ISO2_TO_NAME } from "#country/country"
-import { wordNorm, wordNormLower } from "#normalize"
-import { US_STATE_ABBREVIATIONS, US_STATE_NAMES } from "#us/state"
-
-/**
- * Ambiguous entries printed before the list is truncated.
- */
-const MAX_LISTED_AMBIGUOUS = 12
-
-/**
- * Letters at or below which a token reads as an abbreviation rather than a word.
- */
-const MAX_ABBREVIATION_LETTERS = 3
-
-const BIT = { country_surface: 1, country_ambiguous: 2 }
-const SLOTS = ["country_surface", "country_ambiguous"]
-
-/**
- * Committed output path (a codex-derived artifact, like export-country-surfaces.ts — no argv, so the no-process-globals
- * lint policy holds; codex stays zero-runtime-dep).
- */
-const OUTPUT = repoRootPath("data", "gazetteer", "country-surface-lexicon-v1.json")
-
-/**
- * THE shared word-normalization rule (identical to build-gazetteer-anchor-lexicon.mjs and mirrored in
- * gazetteer_char_paint on both sides): per whitespace-word, strip LEADING/TRAILING characters that are not Unicode
- * letters or digits (keep internal ones: "u.s.a", "timor-leste"), rejoin single-spaced. Entry keys and scanned tokens
- * both pass through it, so "U.S.A." ≡ "u.s.a".
- */
-/**
- * Short alphabetic code (≤3 letters once punctuation is dropped) → exact-uppercase matching.
- */
-const isShortCode = (s: string): boolean => {
-	const letters = s.replaceAll(/[^\p{L}]/gu, "")
-
-	return letters.length > 0 && letters.length <= MAX_ABBREVIATION_LETTERS && /^[\p{L}.\s]+$/u.test(s)
-}
-
-// Homograph set: a single-word country surface that is ALSO a US region (name or abbreviation) reads
-// ambiguously (Georgia the country vs the state, IN = India vs Indiana). Computed from codex so it
-// tracks the US region table, never hand-maintained.
-const usStateNames = new Set(US_STATE_NAMES.map((n) => n.toLowerCase()))
-const usStateAbbrevs = new Set<string>(US_STATE_ABBREVIATIONS as readonly string[])
-
-/**
- * Curated common-word country surfaces — single tokens that appear far more often as ordinary street/venue/locality
- * words than as a trailing country. A SOFT flag (the model still decides), the model-first analogue of Pelias's
- * blacklist (north/south/east/west/street/city/king). Tunable.
- */
-const COMMON_WORD_AMBIGUOUS = new Set(["america", "england", "britain", "turkey", "chad", "jordan", "jersey", "guinea"])
-
-const isAmbiguousName = (lowerKey: string): boolean => usStateNames.has(lowerKey) || COMMON_WORD_AMBIGUOUS.has(lowerKey)
-
-// surface → bits, split across the two match-rule maps (mirrors the gazetteer builder).
-const entries = new Map<string, number>() // lowercase key
-const codeEntries = new Map<string, number>() // exact-uppercase key
-let maxNgram = 1
-
-function add(surface: string): void {
-	const s = surface.trim()
-
-	if (!s) return
-
-	if (isShortCode(s)) {
-		const key = wordNorm(s).toUpperCase()
-
-		if (!key) return
-		// A code that collides with a US-state abbreviation (CA/IN/AL/CO/…) is a homograph → ambiguous.
-		const bits = BIT.country_surface | (usStateAbbrevs.has(key) ? BIT.country_ambiguous : 0)
-		codeEntries.set(key, (codeEntries.get(key) ?? 0) | bits)
-
-		return
-	}
-
-	const key = wordNormLower(s)
-
-	if (!key) return
-	const words = key.split(" ")
-	maxNgram = Math.max(maxNgram, words.length)
-	// Multi-word phrases are unambiguous by construction; single tokens consult the homograph +
-	// common-word rule.
-	const ambiguous = words.length === 1 && isAmbiguousName(key)
-	const bits = BIT.country_surface | (ambiguous ? BIT.country_ambiguous : 0)
-	entries.set(key, (entries.get(key) ?? 0) | bits)
-}
-
-// Curated rich surface forms first (US/GB/DE/… endonyms + abbreviations), then the canonical English
-// name for every remaining ISO 3166-1 alpha-2 — exactly the merge country-surfaces.json performs.
-for (const forms of Object.values(COUNTRY_SURFACE_FORMS)) {
-	for (const f of forms) {
-		add(f)
-	}
-}
-
-for (const [, name] of ISO2_TO_NAME) {
-	add(name)
-}
-
-const ambiguousEntries = [...entries, ...codeEntries].filter(([, b]) => b & BIT.country_ambiguous).map(([k]) => k)
-
-const lexicon = {
-	version: 1,
-	generated_by:
-		"codex/tools/build-country-surface-lexicon.ts (source: @mailwoman/codex COUNTRY_SURFACE_FORMS + ISO2_TO_NAME)",
-	feature_dim: SLOTS.length,
-	slots: SLOTS,
-	bits: BIT,
-	max_ngram: maxNgram,
-	rules: {
-		word_norm:
-			"per whitespace-word: strip leading/trailing chars that are not Unicode letters/digits " +
-			"(keep internal: 'timor-leste', 'u.s.a'); rejoin single-spaced. Applied to BOTH entry keys and scanned tokens.",
-		entries:
-			"case-insensitive; key = word_norm lowercased. country_surface always set; country_ambiguous set for single-token homographs (US region) or curated common-word names.",
-		code_entries:
-			"case-SENSITIVE exact: word_norm(token) == key (keys uppercase; 'in' the word ≠ 'IN' India). n-gram length 1. country_ambiguous set when the code is also a US-state abbreviation.",
-		scan: "longest-first n-gram over whitespace words, left to right, non-overlapping (shared with the gazetteer matcher)",
-		feature:
-			"emitted per-piece row = [country_surface, country_ambiguous] (the raw bits); confidence = 1.0 where country_surface fires.",
-	},
-	entries: Object.fromEntries([...entries].toSorted(([a], [b]) => a.localeCompare(b))),
-	code_entries: Object.fromEntries([...codeEntries].toSorted(([a], [b]) => a.localeCompare(b))),
-}
-
-await makeDirectories(dirname(OUTPUT))
-await writeLocalTextFile(JSON.stringify(lexicon, null, 1) + "\n", OUTPUT)
-
-process.stderr.write(
-	`wrote ${OUTPUT}: ${entries.size} entries + ${codeEntries.size} code_entries, ` +
-		`max_ngram=${maxNgram}, ${ambiguousEntries.length} ambiguous: ${ambiguousEntries.slice(0, MAX_LISTED_AMBIGUOUS).join(", ")}${ambiguousEntries.length > MAX_LISTED_AMBIGUOUS ? ", …" : ""}\n`
-)

package/lib/tools/export-country-surfaces.ts

@@ -1,48 +0,0 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   codex → corpus-python bridge: emit the authoritative country surface forms as JSON so the Python
- *   extract generators can synthesize address tails ("…, USA" / "…, United States of America") without
- *   re-deriving the country name/alias data. `@mailwoman/codex` stays the single source of truth
- *   (COUNTRY_SURFACE_FORMS + ISO2_TO_NAME, salvaged from isp-nexus spatial/countries); this writes a
- *   snapshot the language boundary can't import directly.
- *
- *   Regenerate: `node codex/tools/export-country-surfaces.ts` (writes the corpus-python data file).
- */
-
-import { writeLocalTextFile } from "@mailwoman/core/fs/writers"
-import { repoRootPath } from "@mailwoman/core/paths"
-
-import { COUNTRY_SURFACE_FORMS, ISO2_TO_NAME } from "#country/country"
-
-// Merge: rich surface forms where the codex curates them, else the canonical English name for every
-// ISO 3166-1 alpha-2. Canonical-name-first (the codex's own ordering) so the common form leads.
-const surfaces: Record<string, string[]> = {}
-
-for (const [iso2, forms] of Object.entries(COUNTRY_SURFACE_FORMS)) {
-	surfaces[iso2] = [...forms]
-}
-
-for (const [iso2, name] of ISO2_TO_NAME) {
-	if (!surfaces[iso2]) {
-		surfaces[iso2] = [name]
-	}
-}
-
-const out = repoRootPath("corpus-python", "src", "mailwoman_train", "data", "country-surfaces.json")
-
-await writeLocalTextFile(
-	JSON.stringify(
-		{
-			_generated: "codex/tools/export-country-surfaces.ts from @mailwoman/codex COUNTRY_SURFACE_FORMS + ISO2_TO_NAME",
-			surfaces,
-		},
-		null,
-		2
-	) + "\n",
-	out
-)
-
-process.stderr.write(`wrote ${Object.keys(surfaces).length} countries → ${out}\n`)

package/lib/tools/generate-country-population.ts

@@ -1,142 +0,0 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   Regenerate `codex/country/population.ts` — the per-country population table — from GeoNames
- *   `countryInfo.txt` (https://download.geonames.org/export/dump/countryInfo.txt, CC-BY-4.0). The
- *   output is committed; this tool makes it reproducible (provenance), not a hand-typed dictionary.
- *
- *   Why it exists (#1650): WOF carries no readable population for 147 of 237 country records
- *   (measured against the 2026-08-18 candidate build), so those countries entered every prominence
- *   race at an asserted ZERO — ranked below any namesake hamlet. The magnitude is what a fame race
- *   reads, so a census-vintage figure is entirely sufficient; currency of the estimate is not the
- *   point.
- *
- *   Rows whose GeoNames population is 0 are DROPPED rather than emitted: an entry in this table is a
- *   positive claim, and the consumer's absence branch (`?? undefined`) must stay reachable for
- *   territories GeoNames itself declines to estimate (the meaning-of-zero rule).
- *
- *   Usage: mailwoman dev generate country-population
- */
-
-import { APIClient, pluckResponseData } from "@mailwoman/core/api"
-import { writeLocalTextFile } from "@mailwoman/core/fs/writers"
-import { resolvePackagePath } from "@mailwoman/core/module/resolvers"
-
-const SOURCE = "https://download.geonames.org/export/dump/countryInfo.txt"
-
-/**
- * The committed output path, resolved relative to this module (codex/tools/ → codex/country/). See
- * `generate-country-reference.ts` for why this is `import.meta.url` rather than a `core` path helper (codex is
- * zero-runtime-dep and `core` already references `codex` — importing core here would cycle the project graph).
- */
-const DEFAULT_OUT = resolvePackagePath("@mailwoman/codex", "lib", "country", "population.ts")
-
-/**
- * Tab positions this tool reads from `countryInfo.txt`'s 19-column rows. Named so the parse states which columns it
- * believes in; a GeoNames format change fails the count guard below rather than silently reading the wrong column.
- */
-const COLUMN_ISO2 = 0
-const COLUMN_POPULATION = 7
-const MINIMUM_COLUMNS = 8
-
-/**
- * GeoNames publishes ~250 countries/territories; a parse recovering fewer than this read the wrong column or a
- * truncated body, and the guard fails loudly instead of committing a hollow table.
- */
-const MINIMUM_PLAUSIBLE_COUNTRIES = 200
-
-/**
- * Numbers below five digits are emitted bare — the house numeric-separator style groups by three and only from five
- * digits up (`8450`, not `8_450`).
- */
-const SEPARATOR_MINIMUM = 10_000
-
-/**
- * Options for {@linkcode generateCountryPopulation}.
- */
-export interface GenerateCountryPopulationOptions {
-	/**
-	 * Output path override. Default: `codex/country/population.ts` (the committed table).
-	 */
-	out?: string
-}
-
-/**
- * Summary returned by {@linkcode generateCountryPopulation}.
- */
-export interface GenerateCountryPopulationSummary {
-	countries: number
-	outPath: string
-}
-
-/**
- * Fetch GeoNames `countryInfo.txt` and regenerate the committed `COUNTRY_POPULATION` table.
- */
-export async function generateCountryPopulation(
-	options: GenerateCountryPopulationOptions = {},
-	report?: (line: string) => void
-): Promise<GenerateCountryPopulationSummary> {
-	const outPath = options.out ?? DEFAULT_OUT
-
-	// `responseType: "text"` because the source is a tab-separated dump, not JSON.
-	const text = await new APIClient({ displayName: "geonames-country-info", retry: true })
-		.fetch<string>({ url: SOURCE, responseType: "text" })
-		.then(pluckResponseData)
-
-	const rows: Record<string, number> = {}
-
-	// oxlint-disable-next-line mailwoman/prefer-spliterator -- countryInfo.txt is ~35 kB and bounded (~300 rows)
-	for (const line of text.split("\n")) {
-		if (!line || line.startsWith("#")) continue
-		// oxlint-disable-next-line mailwoman/prefer-spliterator -- one 19-column row
-		const columns = line.split("\t")
-
-		if (columns.length < MINIMUM_COLUMNS) continue
-		const alpha2 = columns[COLUMN_ISO2]!.trim()
-		const population = Number(columns[COLUMN_POPULATION])
-
-		if (!/^[A-Z]{2}$/.test(alpha2)) continue
-
-		if (!Number.isFinite(population) || population <= 0) continue
-
-		rows[alpha2] = population
-	}
-
-	if (Object.keys(rows).length < MINIMUM_PLAUSIBLE_COUNTRIES) {
-		throw new Error(`generateCountryPopulation: only ${Object.keys(rows).length} rows parsed — format drift?`)
-	}
-
-	const body = Object.keys(rows)
-		.toSorted()
-		.map((k) => {
-			const n = rows[k]!
-			const literal = n >= SEPARATOR_MINIMUM ? n.toLocaleString("en-US").replaceAll(",", "_") : String(n)
-
-			return `\t${k}: ${literal},`
-		})
-		.join("\n")
-
-	const header = `/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   GENERATED — do not edit by hand. Per-country population, derived from GeoNames countryInfo.txt
- *   (https://download.geonames.org/export/dump/countryInfo.txt, CC-BY-4.0). Estimates are
- *   census-vintage; a prominence race reads the magnitude, not the currency. Countries GeoNames
- *   declines to estimate are ABSENT, never zero.
- *   Regenerate with: mailwoman dev generate country-population
- */
-
-/**
- * ISO 3166-1 alpha-2 → population estimate.
- */
-export const COUNTRY_POPULATION: Readonly<Record<string, number>> = {`
-
-	await writeLocalTextFile(`${header}\n${body}\n}\n`, outPath)
-	report?.(`wrote ${outPath} (${Object.keys(rows).length} countries)`)
-
-	return { countries: Object.keys(rows).length, outPath }
-}