@mailwoman/corpus 9.2.0 → 9.4.0

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

@@ -0,0 +1,257 @@
+/**
+ * @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 { pathExists, readFileHead, readLocalBuffer, readLocalTextFile } from "@mailwoman/core/fs/readers"
+import { makeDirectories, removePathIfPresent } from "@mailwoman/core/fs/writers"
+import { extractZipEntry } from "@mailwoman/core/fs/zip"
+import { sha256File } from "@mailwoman/core/hash"
+import { join } from "path-ts"
+
+import type { BaseFetchOptions, FetchSummary } from "#tools/fetch/download"
+import { downloadToFile, HTTPStatusError, writeManifest } from "#tools/fetch/download"
+
+/**
+ * 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.
+ *
+ * Walk the string directly rather than constructing a spliterator: the capitals builder already holds whole country
+ * dumps as strings, and a byte-oriented spliterator would UTF-8 encode that input — allocating up to another 350 MB —
+ * merely to inspect its first non-empty line.
+ */
+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
+
+/**
+ * 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[] }> {
+	await makeDirectories(options.outRoot)
+
+	const baseURL = options.baseURL ?? BASE_URL
+	const countryInfoDest = join(options.outRoot, "countryInfo.txt")
+
+	await downloadToFile({
+		url: `${baseURL}/countryInfo.txt`,
+		dest: countryInfoDest,
+		timeoutMs: 120_000,
+		retries: 2,
+		report,
+	})
+
+	const catalog = parseCountryInfo(await readLocalTextFile(countryInfoDest))
+	const countries = options.countries?.map((code) => code.trim().toUpperCase()) ?? catalog.map((row) => row.country)
+
+	const entries: GeonamesDumpFileEntry[] = []
+	const failedCodes: string[] = []
+	const unavailable: string[] = []
+	const skippedPresent: string[] = []
+	const wrongFormatPresent: string[] = []
+	let fetched = 0
+
+	for (const country of countries) {
+		const txtDest = join(options.outRoot, `${country}.txt`)
+
+		if (!options.force && (await pathExists(txtDest))) {
+			if (looksLikeGazetteerDump(await readFileHead(txtDest, FORMAT_SNIFF_BYTES))) {
+				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 removePathIfPresent(zipDest)
+
+			entries.push({
+				country,
+				filename: `${country}.txt`,
+				source_url: url,
+				sha256: await sha256File(txtDest),
+				bytes: (await readLocalBuffer(txtDest)).byteLength,
+			})
+
+			fetched++
+		} catch (error) {
+			await removePathIfPresent(zipDest)
+
+			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/lib/tools/fetch/geonames-postal.ts

@@ -0,0 +1,180 @@
+/**
+ * @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 slice 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 slice 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 slice 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 slice 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 slice 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 { makeDirectories } from "@mailwoman/core/fs/writers"
+import { sha256File } from "@mailwoman/core/hash"
+import { join } from "path-ts"
+
+import type { BaseFetchOptions, FetchSummary } from "#tools/fetch/download"
+import { downloadToFile, HTTPStatusError, writeManifest } from "#tools/fetch/download"
+
+/**
+ * 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)
+	await makeDirectories(destDir)
+
+	const countries = (options.countries ?? GEONAMES_POSTAL_DEFAULT_COUNTRIES).map((code) => code.trim().toUpperCase())
+	const baseURL = options.baseURL ?? BASE_URL
+	const retryDelayMs = options.retryDelayMs
+	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, retryDelayMs, 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 slice 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/lib/tools/fetch/houjin-jp.ts

@@ -0,0 +1,135 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   Fetch the National Tax Agency's corporate-number register (法人番号公表サイト 全件データ): every
+ *   corporation assigned a number, about 5,000,000 rows, with the head-office address as one string
+ *   (国内所在地) plus the prefecture and municipality split out and the postcode. The nationwide CSV
+ *   in Unicode is one zip of about 255 MB. This is the NOISY source for Japanese addresses; the
+ *   Overture rows the JP corpus is built from are the LABEL half.
+ *
+ *   The agency states the three published fields may be used freely by anyone ("どなたでも自由にご利用
+ *   いただくことができます"), with no attribution condition on the download page.
+ *
+ *   The file sits behind a form: the page hands out a per-session token, and a POST with that token,
+ *   `event=download` and the file number of the nationwide Unicode CSV answers the zip. The file
+ *   number is read off the page rather than pinned, because the agency re-issues the files monthly.
+ *
+ *   Invoke via `mailwoman corpus fetch houjin-jp --out-root <path>`.
+ */
+
+import { makeDirectories } from "@mailwoman/core/fs/writers"
+import { sha256File } from "@mailwoman/core/hash"
+import { join } from "path-ts"
+
+import type { BaseFetchOptions, FetchSummary, SourceCollectionManifest } from "#tools/fetch/download"
+import { attachmentFilename, cookieHeader, streamBodyToFile, withRetries, writeManifest } from "#tools/fetch/download"
+
+const SLUG = "houjin-jp"
+const PAGE_URL = "https://www.houjin-bangou.nta.go.jp/download/zenken/"
+const FORM_URL = `${PAGE_URL}index.html`
+const TOKEN_FIELD = "jp.go.nta.houjin_bangou.framework.web.common.CNSFWTokenProcessor.request.token"
+const LICENSE = "国税庁法人番号公表サイト — the three published fields are free for anyone to use (利用規約)"
+const ATTRIBUTION = "国税庁 法人番号公表サイト (National Tax Agency, Corporate Number Publication Site)"
+const USER_AGENT = "Mozilla/5.0 (X11; Linux x86_64) mailwoman-corpus-fetch"
+
+export type FetchHoujinJPOptions = BaseFetchOptions
+
+/**
+ * The nationwide row of the "CSV形式・Unicode" table carries the file number in its `doDownload(N)` handler.
+ */
+function nationwideUnicodeFileNumber(html: string): string | undefined {
+	const start = html.indexOf('id="csv-unicode"')
+
+	if (start === -1) return undefined
+	const section = html.slice(start, html.indexOf('id="xml-unicode"', start))
+	const row = /全国[\s\S]{0,600}?doDownload\((\d+)\)/.exec(section)
+
+	return row?.[1]
+}
+
+export async function fetchHoujinJP(
+	options: FetchHoujinJPOptions,
+	report?: (line: string) => void
+): Promise<FetchSummary> {
+	const destDir = join(options.outRoot, SLUG)
+	await makeDirectories(destDir)
+
+	const page = await fetch(PAGE_URL, { headers: { "user-agent": USER_AGENT, accept: "text/html" } })
+
+	if (!page.ok) throw new Error(`houjin-jp: the download page answered HTTP ${page.status}`)
+	const html = await page.text()
+	const token = new RegExp(`${TOKEN_FIELD.replaceAll(".", "\\.")}" value="([^"]+)"`).exec(html)?.[1]
+	const fileNumber = nationwideUnicodeFileNumber(html)
+
+	if (!token || !fileNumber) {
+		report?.(
+			`  ✗ the page carried ${token ? "a token" : "no token"} and ${fileNumber ? `file ${fileNumber}` : "no nationwide Unicode CSV row"}`
+		)
+
+		return { fetched: 0, skipped: 0, failed: 1, failedCodes: ["zenkoku-csv-unicode"] }
+	}
+
+	report?.(`=== ${SLUG}: nationwide CSV (Unicode), file ${fileNumber}`)
+
+	let filename = `zenkoku_all_${fileNumber}.zip`
+	let bytes: number
+
+	try {
+		bytes = await withRetries(
+			async () => {
+				// A fresh page per attempt: the token is bound to the session cookie, and both may have lapsed.
+				const attemptPage = await fetch(PAGE_URL, { headers: { "user-agent": USER_AGENT, accept: "text/html" } })
+
+				const attemptToken = new RegExp(`${TOKEN_FIELD.replaceAll(".", "\\.")}" value="([^"]+)"`).exec(
+					await attemptPage.text()
+				)?.[1]
+
+				if (!attemptToken) throw new Error("the download page carried no token")
+
+				const res = await fetch(FORM_URL, {
+					method: "POST",
+					headers: {
+						"user-agent": USER_AGENT,
+						cookie: cookieHeader(attemptPage),
+						referer: PAGE_URL,
+						"content-type": "application/x-www-form-urlencoded",
+					},
+					body: new URLSearchParams({ [TOKEN_FIELD]: attemptToken, event: "download", selDlFileNo: fileNumber }),
+					signal: AbortSignal.timeout(1_800_000),
+				})
+
+				if (!res.ok || !(res.headers.get("content-type") ?? "").includes("octet-stream")) {
+					throw new Error(`HTTP ${res.status} ${res.headers.get("content-type") ?? ""}`)
+				}
+
+				filename = attachmentFilename(res, filename)
+
+				return streamBodyToFile(res, join(destDir, filename))
+			},
+			{ report, label: filename }
+		)
+	} catch (error) {
+		report?.(`  ✗ ${(error as Error).message}`)
+
+		return { fetched: 0, skipped: 0, failed: 1, failedCodes: ["zenkoku-csv-unicode"] }
+	}
+
+	const dest = join(destDir, filename)
+	const sha = await sha256File(dest)
+	report?.(`  ✓ ${filename} ${(bytes / 1024 / 1024).toFixed(1)} MB  sha256=${sha}`)
+
+	const manifest: SourceCollectionManifest = {
+		source: SLUG,
+		source_url: PAGE_URL,
+		license: LICENSE,
+		attribution: ATTRIBUTION,
+		downloaded_at: new Date().toISOString(),
+		files: [{ source_url: FORM_URL, downloaded_at: new Date().toISOString(), filename, sha256: sha, bytes }],
+	}
+
+	await writeManifest(join(destDir, "MANIFEST.json"), manifest)
+
+	return { fetched: 1, skipped: 0, failed: 0, failedCodes: [] }
+}

package/lib/tools/fetch/hrsa.ts

@@ -0,0 +1,59 @@
+/**
+ * @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 { makeDirectories } from "@mailwoman/core/fs/writers"
+import { sha256File } from "@mailwoman/core/hash"
+import { join } from "path-ts"
+
+import type { BaseFetchOptions, FetchSummary, SourceManifest } from "#tools/fetch/download"
+import { downloadToFile, writeManifest } from "#tools/fetch/download"
+
+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
+
+export async function fetchHRSA(options: FetchHRSAOptions, report?: (line: string) => void): Promise<FetchSummary> {
+	const destDir = join(options.outRoot, SLUG)
+	await makeDirectories(destDir)
+	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: [] }
+}