@mailwoman/corpus 9.2.0 → 9.4.0

package/src/tools/fetch/geonames-dump.ts

@@ -1,267 +0,0 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   Fetch GeoNames per-country GAZETTEER dumps — the 19-column `<CC>.txt` files under
- *   `https://download.geonames.org/export/dump/` (NOT the postal exports; those are `export/zip/` and
- *   `geonames-postal.ts`'s job). The dumps carry feature classes and codes (column 8: `PPLC` national capital,
- *   `PPLA` first-order administrative seat), which is what the capitals reference build consumes (#1880).
- *
- *   The catalog question is answered by the SOURCE, not by an ISO list: `countryInfo.txt` in the same directory
- *   enumerates every country GeoNames publishes, one row per ISO alpha-2 code, and also names each country's
- *   capital — the cross-check the capitals build grades its `PPLC` extraction against. Fetch that first; derive
- *   the country set from it; then a dump absent from disk is a measured gap against the source's own catalog
- *   rather than a silent hole. The dump directory may hold files this tool did not fetch: present files are never
- *   overwritten, and a present `<CC>.txt` that is NOT a 19-column gazetteer dump (GeoNames' postal exports share
- *   the basename) is reported as `wrong_format_present`, never counted as coverage.
- */
-
-import { existsSync, mkdirSync } from "node:fs"
-import { open, readFile, rm } from "node:fs/promises"
-import { join } from "node:path"
-
-import { extractZipEntry } from "@mailwoman/core/fs/zip"
-import { sha256File } from "@mailwoman/core/utils"
-
-import type { BaseFetchOptions, FetchSummary } from "./download.ts"
-import { downloadToFile, HTTPStatusError, writeManifest } from "./download.ts"
-
-/**
- * The one status that means "the source does not publish this country" rather than "the transfer failed".
- */
-const HTTP_NOT_FOUND = 404
-
-const SLUG = "geonames-dump"
-
-/**
- * GeoNames' gazetteer-dump directory — one zip per ISO alpha-2 code holding `<CC>.txt`, plus `countryInfo.txt` as a
- * bare text file.
- */
-const BASE_URL = "https://download.geonames.org/export/dump"
-
-export interface FetchGeonamesDumpOptions extends BaseFetchOptions {
-	/**
-	 * ISO alpha-2 codes, any casing. Absent → every country `countryInfo.txt` enumerates.
-	 */
-	countries?: readonly string[]
-	/**
-	 * Dump directory to read from. Defaults to GeoNames' own; exists so the 404 and coverage behaviour can be exercised
-	 * against a local server.
-	 */
-	baseURL?: string
-	/**
-	 * Refetch a dump whose `<CC>.txt` already exists. Default false — the tool fills gaps.
-	 */
-	force?: boolean
-}
-
-interface GeonamesDumpFileEntry {
-	country: string
-	filename: string
-	source_url: string
-	sha256: string
-	bytes: number
-}
-
-export interface GeonamesDumpManifest {
-	source: string
-	base_url: string
-	license: string
-	attribution: string
-	downloaded_at: string
-	files: GeonamesDumpFileEntry[]
-	/**
-	 * `<CC>.txt` files already on disk and left alone — the hand-fetched population this tool extends.
-	 */
-	skipped_present: string[]
-	/**
-	 * Countries in the source catalog that the source's dump directory nonetheless 404s — a fact about the source,
-	 * recorded so a later reader does not spend the fetch to rediscover it.
-	 */
-	unavailable: string[]
-	/**
-	 * Present `<CC>.txt` files that are NOT 19-column gazetteer dumps — GeoNames' postal exports share the same basename,
-	 * and seven tier-1 postal files sat at these paths reading as "present" until the capitals build found them
-	 * capital-less. Left in place (this tool never overwrites data it did not fetch); the fix is to move the file to its
-	 * own home and rerun.
-	 */
-	wrong_format_present: string[]
-}
-
-/**
- * Column count of a gazetteer dump row — the discriminator against GeoNames' 12-column postal exports, which share the
- * `<CC>.txt` basename.
- */
-const GAZETTEER_DUMP_COLUMNS = 19
-
-/**
- * True when the first non-empty line carries the gazetteer dump's 19 tab-separated columns. Accepts a partial head read
- * — the first line is the whole question, so callers need not hand it a resident 350 MB dump.
- */
-export function looksLikeGazetteerDump(text: string): boolean {
-	let start = 0
-
-	while (start < text.length) {
-		const end = text.indexOf("\n", start)
-		const line = end === -1 ? text.slice(start) : text.slice(start, end)
-
-		if (line.trim()) {
-			let tabs = 0
-
-			for (let i = line.indexOf("\t"); i !== -1; i = line.indexOf("\t", i + 1)) {
-				tabs++
-			}
-
-			return tabs === GAZETTEER_DUMP_COLUMNS - 1
-		}
-
-		if (end === -1) break
-
-		start = end + 1
-	}
-
-	return false
-}
-
-/**
- * Parse the ISO codes (column 1) and capital names (column 6) out of `countryInfo.txt` — `#`-prefixed lines are the
- * file's own commentary.
- */
-export function parseCountryInfo(text: string): Array<{ country: string; capital: string }> {
-	const rows: Array<{ country: string; capital: string }> = []
-
-	// oxlint-disable-next-line mailwoman/prefer-spliterator -- countryInfo.txt is ~40 KB, one row per country on Earth; it cannot grow past that
-	for (const line of text.split("\n")) {
-		if (line.startsWith("#") || !line.trim()) continue
-
-		// oxlint-disable-next-line mailwoman/prefer-spliterator -- one bounded 19-column catalog row
-		const cols = line.split("\t")
-		const country = cols[0]?.trim().toUpperCase()
-
-		if (country?.length === 2) {
-			rows.push({ country, capital: cols[5]?.trim() ?? "" })
-		}
-	}
-
-	return rows
-}
-
-/**
- * Bytes read to classify a present file: enough to cover a first dump row whose alternate-names column runs long (they
- * reach several KB), a fraction of the largest dumps (US.txt is ~350 MB).
- */
-const FORMAT_SNIFF_BYTES = 65_536
-
-async function readFileHead(path: string): Promise<string> {
-	const handle = await open(path, "r")
-
-	try {
-		const buffer = Buffer.alloc(FORMAT_SNIFF_BYTES)
-		const { bytesRead } = await handle.read(buffer, 0, FORMAT_SNIFF_BYTES, 0)
-
-		return buffer.subarray(0, bytesRead).toString("utf8")
-	} finally {
-		await handle.close()
-	}
-}
-
-/**
- * Download `countryInfo.txt` plus every missing `<CC>.zip`, extracting each to `<outRoot>/<CC>.txt` beside the
- * hand-fetched dumps, with a `MANIFEST.json` naming fetched, skipped-present, and source-unavailable countries.
- */
-export async function fetchGeonamesDumps(
-	options: FetchGeonamesDumpOptions,
-	report?: (line: string) => void
-): Promise<FetchSummary & { skippedPresent: string[] }> {
-	mkdirSync(options.outRoot, { recursive: true })
-
-	const baseURL = options.baseURL ?? BASE_URL
-	const countryInfoDest = join(options.outRoot, "countryInfo.txt")
-
-	await downloadToFile({
-		url: `${baseURL}/countryInfo.txt`,
-		dest: countryInfoDest,
-		timeoutMs: 120_000,
-		retries: 2,
-		report,
-	})
-
-	const catalog = parseCountryInfo(await readFile(countryInfoDest, "utf8"))
-	const countries = options.countries?.map((code) => code.trim().toUpperCase()) ?? catalog.map((row) => row.country)
-
-	const entries: GeonamesDumpFileEntry[] = []
-	const failedCodes: string[] = []
-	const unavailable: string[] = []
-	const skippedPresent: string[] = []
-	const wrongFormatPresent: string[] = []
-	let fetched = 0
-
-	for (const country of countries) {
-		const txtDest = join(options.outRoot, `${country}.txt`)
-
-		if (!options.force && existsSync(txtDest)) {
-			if (looksLikeGazetteerDump(await readFileHead(txtDest))) {
-				skippedPresent.push(country)
-			} else {
-				report?.(`✗ ${country}.txt is present but is not a 19-column gazetteer dump — move it aside and rerun`)
-				wrongFormatPresent.push(country)
-			}
-
-			continue
-		}
-
-		const filename = `${country}.zip`
-		const url = `${baseURL}/${filename}`
-		const zipDest = join(options.outRoot, filename)
-
-		report?.(`=== ${SLUG} / ${country}`)
-
-		try {
-			await downloadToFile({ url, dest: zipDest, timeoutMs: 300_000, retries: 2, report })
-			await extractZipEntry(zipDest, `${country}.txt`, txtDest)
-			await rm(zipDest, { force: true })
-
-			entries.push({
-				country,
-				filename: `${country}.txt`,
-				source_url: url,
-				sha256: await sha256File(txtDest),
-				bytes: (await readFile(txtDest)).byteLength,
-			})
-
-			fetched++
-		} catch (error) {
-			await rm(zipDest, { force: true })
-
-			const message = error instanceof Error ? error.message : String(error)
-
-			// Branch on the TYPED status (the geonames-postal lesson): message prose contains the URL, and a URL
-			// can contain any substring.
-			if (error instanceof HTTPStatusError && error.status === HTTP_NOT_FOUND) {
-				report?.(`✗ ${country}: GeoNames publishes no gazetteer dump for this country`)
-				unavailable.push(country)
-			} else {
-				report?.(`✗ ${country}: ${message}`)
-			}
-
-			failedCodes.push(country)
-		}
-	}
-
-	const manifest: GeonamesDumpManifest = {
-		source: SLUG,
-		base_url: baseURL,
-		license: "CC-BY-4.0",
-		attribution: "GeoNames",
-		downloaded_at: new Date().toISOString(),
-		files: entries,
-		skipped_present: skippedPresent.toSorted(),
-		unavailable,
-		wrong_format_present: wrongFormatPresent.toSorted(),
-	}
-
-	await writeManifest(join(options.outRoot, "MANIFEST.json"), manifest)
-
-	return { fetched, skipped: skippedPresent.length, failed: failedCodes.length, failedCodes, skippedPresent }
-}

package/src/tools/fetch/geonames-postal.ts

@@ -1,180 +0,0 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   Fetch the GeoNames per-country postal-code exports — the only source in this family that carries a
- *   `(postcode, locality, region)` triple with the NAMES inline.
- *
- *   Source : https://download.geonames.org/export/zip/<CC>.zip
- *   License: CC-BY-4.0, attribute "GeoNames". Tier B.
- *
- *   ## Why this source and not a join
- *
- *   A postcode shard is only useful to the corpus if a postcode reaches a locality and a region. Two routes exist and
- *   only one of them works everywhere:
- *
- *   - **`parent_id`** — `postalcode-intl.db` carries a real parent that resolves in the admin gazetteer (measured: NL
- *     97.5% of rows linked, FR 90.7%, DE 66.1%, ES 34.9%, IT 27.4%, and 93.8–100% of those resolve to a
- *     `locality`/`localadmin`). It is also the ONLY shard that does: `postalcode-geonames-intl.db` and every
- *     `postalcode-<cc>-overture.db` carry `parent_id = 0` on every row, so five countries have this route and the rest
- *     have none.
- *   - **Nearest locality centroid** — the obvious fallback, and it does not work. Scored against the `parent_id` truth
- *     on 1,000–1,500 postcodes per country: NL 81.1%, DE 44.7%, ES 35.1%, FR 29.0%, IT 13.5%. A locality's centroid
- *     sits at its middle, so a postcode near the edge is routinely closer to a neighbouring town's centroid than to its
- *     own. Under 50% for four of the five, which is not a join.
- *
- *   The GeoNames export sidesteps both: columns 3 and 4 ARE the place and admin1 names, so there is nothing to join and
- *   nothing to approximate. `@mailwoman/corpus`'s `geonames-postal` adapter consumes it directly.
- *
- *   ## Coverage is not universal, and the gap is the point
- *
- *   GeoNames publishes ~80 countries, NOT all of them. Venezuela returns 404 — so a VE postcode shard cannot be built
- *   from this source at any effort, and that is an acquisition question rather than a build one. Ask for a country
- *   before assuming it is there; an absent country fails as one entry, never as the whole run.
- *
- *   ## Row counts from this source overstate, for some countries by exactly 2×
- *
- *   Countries whose postcode format contains a hyphen are published TWICE — once `3750-000`, once `3750000`. Measured
- *   on the shard built from this source: PT 395,544 rows over 197,772 distinct codes and PL 40,598 over 20,299, both
- *   exactly 2.00×, while AU, CZ and AT (no hyphen in the format) are 1.00×. A consumer sizing a shard from the row
- *   count doubles its estimate for those countries.
- *
- *   ## Why `downloadToFile` and not `APIClient`
- *
- *   Same split `AGENTS.md` draws and the `ourairports` sibling explains: these are static file transfers from a plain
- *   file host, run once per refresh. The pacing, retry and caching `APIClient` exists for have nothing to act on here.
- *
- *   Invoke via `mailwoman corpus fetch geonames-postal --countries pt,au,nz`.
- */
-
-import { mkdirSync } from "node:fs"
-import { join } from "node:path"
-
-import { sha256File } from "@mailwoman/core/utils"
-
-import type { BaseFetchOptions, FetchSummary } from "./download.ts"
-import { downloadToFile, HTTPStatusError, writeManifest } from "./download.ts"
-
-/**
- * The one status that means "the source does not publish this country" rather than "the transfer failed".
- */
-const HTTP_NOT_FOUND = 404
-
-const SLUG = "geonames-postal"
-
-/**
- * GeoNames' own export directory. One zip per ISO alpha-2 code, each holding `<CC>.txt` plus the shared `readme.txt`.
- */
-const BASE_URL = "https://download.geonames.org/export/zip"
-
-/**
- * Countries fetched when the caller names none.
- *
- * These are the ones the corpus wants and cannot get from `postalcode-intl.db`'s `parent_id` route — see the header for
- * why that route covers exactly five countries. Venezuela is deliberately absent because GeoNames does not publish it.
- */
-export const GEONAMES_POSTAL_DEFAULT_COUNTRIES = ["PT", "AU", "NZ", "IE", "BR", "ZA", "MX"] as const
-
-export interface FetchGeonamesPostalOptions extends BaseFetchOptions {
-	/**
-	 * ISO alpha-2 codes, in any casing. Defaults to {@linkcode GEONAMES_POSTAL_DEFAULT_COUNTRIES}.
-	 */
-	countries?: readonly string[]
-	/**
-	 * Export directory to read from. Defaults to GeoNames' own. Exists so the 404-is-a-coverage-finding behaviour can be
-	 * exercised against a local server rather than by asking GeoNames for a country it does not have.
-	 */
-	baseURL?: string
-}
-
-interface GeonamesPostalFileEntry {
-	country: string
-	filename: string
-	source_url: string
-	sha256: string
-	bytes: number
-}
-
-interface GeonamesPostalManifest {
-	source: string
-	base_url: string
-	license: string
-	attribution: string
-	downloaded_at: string
-	files: GeonamesPostalFileEntry[]
-	/**
-	 * Countries asked for and NOT published by GeoNames, recorded so a later reader does not spend the fetch again to
-	 * rediscover it. An absence here is a fact about the source, not about the run.
-	 */
-	unavailable: string[]
-}
-
-/**
- * Download the requested GeoNames postal zips into `<outRoot>/geonames-postal/`, with a sibling `MANIFEST.json`
- * carrying each file's origin URL, sha256 and byte count, plus the countries the source does not publish.
- *
- * A country the source does not carry is counted as failed and named in `failedCodes` — it does not stop the rest.
- */
-export async function fetchGeonamesPostal(
-	options: FetchGeonamesPostalOptions,
-	report?: (line: string) => void
-): Promise<FetchSummary> {
-	const destDir = join(options.outRoot, SLUG)
-	mkdirSync(destDir, { recursive: true })
-
-	const countries = (options.countries ?? GEONAMES_POSTAL_DEFAULT_COUNTRIES).map((code) => code.trim().toUpperCase())
-	const baseURL = options.baseURL ?? BASE_URL
-	const entries: GeonamesPostalFileEntry[] = []
-	const failedCodes: string[] = []
-	const unavailable: string[] = []
-	let fetched = 0
-	let failed = 0
-
-	for (const country of countries) {
-		const filename = `${country}.zip`
-		const url = `${baseURL}/${filename}`
-		const dest = join(destDir, filename)
-
-		report?.(`=== ${SLUG} / ${country}`)
-
-		try {
-			const { bytes } = await downloadToFile({ url, dest, timeoutMs: 300_000, retries: 2, report })
-
-			entries.push({ country, filename, source_url: url, sha256: await sha256File(dest), bytes })
-
-			fetched++
-		} catch (error) {
-			const message = error instanceof Error ? error.message : String(error)
-
-			// A 404 here means GeoNames does not publish the country at all, which is a different finding from a failed
-			// transfer and the one a caller planning a shard needs to see. Branch on the TYPED status: matching message
-			// prose classified a 500 as "unpublished" whenever the URL happened to contain the substring 404 — an
-			// ephemeral test-server port did exactly that in CI.
-			if (error instanceof HTTPStatusError && error.status === HTTP_NOT_FOUND) {
-				report?.(`✗ ${country}: GeoNames does not publish a postal export for this country`)
-				unavailable.push(country)
-			} else {
-				report?.(`✗ ${country}: ${message}`)
-			}
-
-			failedCodes.push(country)
-
-			failed++
-		}
-	}
-
-	const manifest: GeonamesPostalManifest = {
-		source: "GeoNames postal codes",
-		base_url: baseURL,
-		license: "CC-BY-4.0",
-		attribution: "GeoNames",
-		downloaded_at: new Date().toISOString(),
-		files: entries,
-		unavailable,
-	}
-
-	await writeManifest(join(destDir, "MANIFEST.json"), manifest)
-
-	return { fetched, skipped: 0, failed, failedCodes }
-}

package/src/tools/fetch/hrsa.ts

@@ -1,68 +0,0 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   Re-fetch the HRSA Health Center Service Delivery Sites CSV. Source for the `usgov-hrsa-fqhc`
- *   adapter. US Public Domain.
- *
- *   Uses Node's built-in fetch (gzip/brotli) and streaming sha256 instead of curl + sha256sum, and
- *   writes the same sibling `MANIFEST.json` (origin URL + fetch timestamp + byte count + sha256) so
- *   downstream adapters can verify provenance.
- *
- *   Invoke via `mailwoman corpus fetch hrsa --out-root <path>`.
- */
-
-import { mkdirSync } from "node:fs"
-import { join } from "node:path"
-
-import { sha256File } from "@mailwoman/core/utils"
-
-import type { BaseFetchOptions, FetchSummary } from "./download.ts"
-import { downloadToFile, writeManifest } from "./download.ts"
-
-const SLUG = "usgov-hrsa-fqhc"
-const FILENAME = "Health_Center_Service_Delivery_and_LookAlike_Sites.csv"
-const SOURCE_URL = `https://data.hrsa.gov/DataDownload/DD_Files/${FILENAME}`
-
-export type FetchHRSAOptions = BaseFetchOptions
-
-interface SourceManifest {
-	source_url: string
-	downloaded_at: string
-	filename: string
-	sha256: string
-	bytes: number
-}
-
-export async function fetchHRSA(options: FetchHRSAOptions, report?: (line: string) => void): Promise<FetchSummary> {
-	const destDir = join(options.outRoot, SLUG)
-	mkdirSync(destDir, { recursive: true })
-	const dest = join(destDir, FILENAME)
-
-	report?.(`=== ${SLUG} / ${FILENAME}`)
-
-	const { bytes } = await downloadToFile({
-		url: SOURCE_URL,
-		dest,
-		timeoutMs: 600_000,
-		headers: { "Accept-Encoding": "gzip, br" },
-		report,
-	})
-
-	const sha = await sha256File(dest)
-
-	const manifest: SourceManifest = {
-		source_url: SOURCE_URL,
-		downloaded_at: new Date().toISOString(),
-		filename: FILENAME,
-		sha256: sha,
-		bytes,
-	}
-
-	await writeManifest(join(destDir, "MANIFEST.json"), manifest)
-
-	report?.(`  ✓ ${(bytes / 1024 / 1024).toFixed(1)} MB  sha256=${sha}`)
-
-	return { fetched: 1, skipped: 0, failed: 0, failedCodes: [] }
-}

package/src/tools/fetch/imls-pls.ts

@@ -1,158 +0,0 @@
-/**
- * @copyright Sister Software
- * @license AGPL-3.0
- * @author Teffen Ellis, et al.
- *
- *   Re-fetch the IMLS Public Libraries Survey (PLS) outlet-level data. Each US public library branch
- *   (outlet) is one row, ~17K rows with address fields. Source for the `usgov-imls-pls` adapter. US
- *   Public Domain (federal statistical survey).
- *
- *   The FY 2023 release is the most current as of 2026-05. IMLS ships a single ZIP containing CSV,
- *   SAS, and SPSS variants. We extract the outlet-level CSV (pls_fy*_outlet*.csv or similar) and
- *   discard the rest. The administrative-entity (system-level) CSV is intentionally skipped — it has
- *   no per-branch address detail.
- *
- *   Uses Node's built-in fetch (gzip/brotli) and streaming sha256 instead of curl + sha256sum. The
- *   ZIP is unpacked with the `unzip` binary via `node:child_process` (no clean Node equivalent for
- *   member listing + selective extraction).
- *
- *   Invoke via `mailwoman corpus fetch imls-pls --out-root <path>`. Idempotent: if dest CSV exists
- *   and sha matches MANIFEST, skips download.
- */
-
-/* oxlint-disable sister-software/prefer-region-over-marks -- these markers label steps inside one
-   procedure, not sections of declarations. A region there folds nothing a reader wants folded. */
-
-import { existsSync, mkdirSync, statSync } from "node:fs"
-import { rm } from "node:fs/promises"
-import { basename, join } from "node:path"
-
-import { BYTES_PER_KIB, ByteFormatter } from "@mailwoman/core/fs/utils"
-import { extractZipEntry, listZipEntries } from "@mailwoman/core/fs/zip"
-import { sha256File } from "@mailwoman/core/utils"
-
-import type { BaseFetchOptions, FetchSummary } from "./download.ts"
-import { downloadToFile, readManifest, writeManifest } from "./download.ts"
-
-/**
- * Bytes per KiB — the divisor for human-readable sizes, and the floor below which a "download" is an error page rather
- * than data.
- */
-
-/**
- * The PLS FY 2023 bulk CSV ZIP (most recent as of 2026-05). If IMLS publishes a newer year, update this URL.
- */
-const ZIP_URL = "https://www.imls.gov/sites/default/files/2025-08/pls_fy2023_csv.zip"
-const SLUG = "usgov-imls-pls"
-
-export type FetchIMLSPLSOptions = BaseFetchOptions
-
-interface SourceManifest {
-	source_url: string
-	downloaded_at: string
-	filename: string
-	sha256: string
-	bytes: number
-}
-
-export async function fetchIMLSPLS(
-	options: FetchIMLSPLSOptions,
-	report?: (line: string) => void
-): Promise<FetchSummary> {
-	const destDir = join(options.outRoot, SLUG)
-	mkdirSync(destDir, { recursive: true })
-
-	const zipDest = join(destDir, basename(ZIP_URL))
-	const manifestPath = join(destDir, "MANIFEST.json")
-
-	report?.(`=== ${SLUG}`)
-
-	// MARK: Idempotency check
-
-	const recorded = await readManifest<Partial<SourceManifest>>(manifestPath)
-
-	if (recorded?.sha256 && recorded.filename) {
-		const recordedPath = join(destDir, recorded.filename)
-
-		if (existsSync(recordedPath) && (await sha256File(recordedPath)) === recorded.sha256) {
-			report?.("  ✓ Already current (sha256 matches MANIFEST) — skipping download.")
-
-			return { fetched: 0, skipped: 1, failed: 0, failedCodes: [] }
-		}
-	}
-
-	// MARK: Download ZIP
-
-	report?.(`  Downloading ${ZIP_URL} ...`)
-
-	const { bytes: zipSize } = await downloadToFile({
-		url: ZIP_URL,
-		dest: zipDest,
-		timeoutMs: 600_000,
-		headers: { "Accept-Encoding": "gzip, br" },
-		report,
-	})
-
-	report?.(`  Downloaded: ${ByteFormatter.formatIEC(zipSize)}`)
-
-	if (zipSize < BYTES_PER_KIB) {
-		report?.(`  ✗ Response too small (${zipSize} bytes) — probable error page`)
-
-		return { fetched: 0, skipped: 0, failed: 1, failedCodes: [SLUG] }
-	}
-
-	// Discover the outlet-level CSV inside the ZIP.
-	// Outlet files match: pls_fy*outlet*.csv (case-insensitive)
-	// Administrative-entity files match: pls_fy*ae*.csv — we skip those.
-	report?.("  Inspecting ZIP contents ...")
-	const entries = (await listZipEntries(zipDest)).map((entry) => entry.name)
-
-	let csvName = entries.find((name) => /pls_fy.*outlet.*\.csv/i.test(name))
-
-	// Fallback: if IMLS renames the file, grab any CSV that is NOT the ae file.
-	if (!csvName) {
-		csvName = entries.find((name) => /\.csv$/i.test(name) && !/system|state|_ae\b|_se\b/i.test(name))
-	}
-
-	if (!csvName) {
-		report?.("  Available files in ZIP:")
-
-		for (const name of entries) {
-			report?.(`    ${name}`)
-		}
-
-		report?.("  ✗ Could not identify outlet CSV — inspect above listing and update this module")
-
-		return { fetched: 0, skipped: 0, failed: 1, failedCodes: [SLUG] }
-	}
-
-	report?.(`  Extracting outlet CSV: ${csvName}`)
-
-	const csvDest = join(destDir, basename(csvName))
-
-	await extractZipEntry(zipDest, csvName, csvDest)
-	const csvSize = statSync(csvDest).size
-	const csvSha = await sha256File(csvDest)
-
-	// MARK: Remove ZIP (small, but keep destDir clean)
-
-	await rm(zipDest, { force: true })
-	report?.("  Removed ZIP (CSV kept)")
-
-	// MARK: Write MANIFEST
-
-	const manifest: SourceManifest = {
-		source_url: ZIP_URL,
-		downloaded_at: new Date().toISOString(),
-		filename: basename(csvName),
-		sha256: csvSha,
-		bytes: csvSize,
-	}
-
-	await writeManifest(manifestPath, manifest)
-
-	report?.(`  ✓ ${(csvSize / 1024 / 1024).toFixed(1)} MB  sha256=${csvSha}`)
-	report?.(`  MANIFEST written to ${manifestPath}`)
-
-	return { fetched: 1, skipped: 0, failed: 0, failedCodes: [] }
-}