@mailwoman/corpus 9.3.0 → 9.4.0

package/lib/recipes/anchor-absorption.ts

@@ -14,12 +14,12 @@
 
 import { dataRootPath } from "@mailwoman/core/data-root"
 import { readLocalJSONFile } from "@mailwoman/core/fs/readers"
+import { makeLcg } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 
-import { makeLcg, sliceSourceID, type CorpusRecipe } from "#recipes/scaffold"
+import { sliceSourceID, type CorpusRecipe } from "#recipes/scaffold"
 import { synthesizeAnchorAbsorptionRow } from "#synthesizers/anchor-absorption"
 import { alignRow } from "#utils"
-
 /**
  * The leading-5-digit source: the real US ZIPs in the postcode anchor's pilot lookup. Resolved through the data-root
  * helper (the lab default is `$MAILWOMAN_DATA_ROOT`), never re-hardcoded.

package/lib/recipes/bare-country.ts

@@ -18,9 +18,9 @@
  */
 
 import { COUNTRY_SURFACE_FORMS, CountryNames, matchCountry } from "@mailwoman/codex/country"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 
-import { alignAndWrite, makeMulberry32, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
-
+import { alignAndWrite, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 /**
  * Surfaces shorter than this are the code register (`JP`, `GER`), not a name — excluded (see the module doc).
  */

package/lib/recipes/boundary-stress.ts

@@ -12,7 +12,9 @@
  *   `synthesizeBoundaryStressRow` is NOT re-exported from the corpus index — imported directly here.
  */
 
-import { makeMulberry32, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
+
+import { type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 import { type BoundaryStressTemplate, synthesizeBoundaryStressRow } from "#synthesizers/boundary-stress"
 import { alignRow } from "#utils"
 

package/lib/recipes/country-balanced.ts

@@ -28,10 +28,11 @@ import { COUNTRY_SURFACE_FORMS, CountryNames } from "@mailwoman/codex/country"
 import { dataRootPath } from "@mailwoman/core/data-root"
 import { isPresent } from "@mailwoman/core/objects"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, readOATuples, type CorpusRecipe } from "#recipes/scaffold"
+import { readOATuples, type CorpusRecipe } from "#recipes/scaffold"
 import { pick } from "#synthesizers/utils"
 import type { CanonicalRow } from "#types"
 import { alignRow } from "#utils"

package/lib/recipes/cz-pcfirst-preposition.ts

@@ -14,8 +14,9 @@
  *   cycle (the lesson); canonical and city-first keep the polarity balanced (the v1.9.9 lesson).
  */
 
-import { alignAndWrite, makeMulberry32, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 
+import { alignAndWrite, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 /**
  * The order-cycle slot for the STREET-LESS form (`«city» «pc», Česko`) — the exact surface of the
  * `cz-full-praha-100-00` board row, whose absence from the street-bearing orders was the v4.5.0 no-promote's measured

package/lib/recipes/fr-admin-split.ts

@@ -33,10 +33,11 @@
 import { departementForCodePostal } from "@mailwoman/codex/fr"
 import { tempRootPath } from "@mailwoman/core/data-root"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import { CSVSpliterator, Delimiters } from "spliterator"
 
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, type CorpusRecipe } from "#recipes/scaffold"
+import type { CorpusRecipe } from "#recipes/scaffold"
 import type { CanonicalRow } from "#types"
 import { alignRow } from "#utils"
 

package/lib/recipes/fr-bare-street.ts

@@ -21,9 +21,10 @@
  */
 
 import { FR_VOIE_TYPES } from "@mailwoman/codex/fr"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 
 import { decomposeFrStreet } from "#adapters/ban/street-decompose"
-import { alignAndWrite, makeMulberry32, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
+import { alignAndWrite, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 
 /**
  * Canonical voie type (lowercase, accent-kept) → its most common written abbreviation, from the codex table's first

package/lib/recipes/fr-fragment.ts

@@ -68,10 +68,11 @@
  *   blow-up). Disable that mask for any run including this slice.
  */
 
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import { TextSpliterator } from "spliterator"
 
 import { decomposeFrStreet } from "#adapters/ban/street-decompose"
-import { alignAndWrite, makeMulberry32, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
+import { alignAndWrite, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 import { pick } from "#synthesizers/utils"
 
 /**

package/lib/recipes/fr-lieudit.ts

@@ -30,15 +30,15 @@ import { dataRootPath } from "@mailwoman/core/data-root"
 import { readDirectory } from "@mailwoman/core/fs/readers"
 import { isPresent } from "@mailwoman/core/objects"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import { join, type PathBuilderLike } from "path-ts"
 
 import { decomposeFrStreet } from "#adapters/ban/street-decompose"
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, type CorpusRecipe } from "#recipes/scaffold"
+import type { CorpusRecipe } from "#recipes/scaffold"
 import { pick } from "#synthesizers/utils"
 import type { CanonicalRow } from "#types"
 import { alignRow } from "#utils"
-
 /**
  * Matches the `ban` adapter's Tier-B election for BAN data.
  */

package/lib/recipes/fr-order.ts

@@ -33,9 +33,10 @@
 import { dataRootPath } from "@mailwoman/core/data-root"
 import { isPresent } from "@mailwoman/core/objects"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, readOATuples, type CorpusRecipe } from "#recipes/scaffold"
+import { readOATuples, type CorpusRecipe } from "#recipes/scaffold"
 import { pick } from "#synthesizers/utils"
 import type { CanonicalRow } from "#types"
 import { alignRow } from "#utils"

package/lib/recipes/german.ts

@@ -27,12 +27,13 @@
 
 import { dataRootPath } from "@mailwoman/core/data-root"
 import { pathExists } from "@mailwoman/core/fs/readers"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { WOFDatabase } from "@mailwoman/resolver-wof-sqlite/schema"
 import { DatabaseClient } from "@mailwoman/sqlite/client"
 import type { PathBuilderLike } from "path-ts"
 
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, readZippedCSVRecords, type CorpusRecipe } from "#recipes/scaffold"
+import { readZippedCSVRecords, type CorpusRecipe } from "#recipes/scaffold"
 import { synthesizeGermanRow, type LocaleBaseTuple } from "#synthesizers/german"
 import { alignRow } from "#utils"
 

package/lib/recipes/house-venue.ts

@@ -10,9 +10,10 @@
  *   scripts/build-house-venue-slice.mjs.
  */
 
+import { makeLcg } from "@mailwoman/core/utils"
+
 import {
 	alignAndWrite,
-	makeLcg,
 	readTuples,
 	sliceSourceID,
 	SYNTHETIC_TUPLE_LICENSE as LICENSE,

package/lib/recipes/intersection.ts

@@ -40,11 +40,12 @@ import { pathExists } from "@mailwoman/core/fs/readers"
 import { writeLocalJSONFile } from "@mailwoman/core/fs/writers"
 import { repoRootPath } from "@mailwoman/core/paths"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 import { JSONSpliterator } from "spliterator"
 
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, readZippedCSVRecords, type CorpusRecipe } from "#recipes/scaffold"
+import { readZippedCSVRecords, type CorpusRecipe } from "#recipes/scaffold"
 import { pick, weightedPick } from "#synthesizers/utils"
 import type { CanonicalRow, LabeledRow } from "#types"
 import { alignRow } from "#utils"

package/lib/recipes/locale.ts

@@ -30,11 +30,12 @@ import { dataRootPath } from "@mailwoman/core/data-root"
 import { openReadStream } from "@mailwoman/core/fs/streams"
 import { readZipEntry } from "@mailwoman/core/fs/zip"
 import { isPresent } from "@mailwoman/core/objects"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 import { CSVSpliterator } from "spliterator"
 
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, type CorpusRecipe } from "#recipes/scaffold"
+import type { CorpusRecipe } from "#recipes/scaffold"
 import { type LocaleBaseTuple, type SynthesizedLocaleRow, synthesizeLocaleRow } from "#synthesizers/german"
 import { pick } from "#synthesizers/utils"
 import { alignRow } from "#utils"

package/lib/recipes/nl-postcode.ts

@@ -18,7 +18,9 @@
  *   keep polarity balanced (the v1.9.9 lesson).
  */
 
-import { alignAndWrite, makeMulberry32, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
+
+import { alignAndWrite, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 
 /**
  * "1012LG" → "1012 LG". The tuples carry the unspaced OA form; the spaced form is the failing case.

package/lib/recipes/no-fragment.ts

@@ -35,16 +35,10 @@
  *   list. Diacritic-KEEPING normalizer, matching the board (see the norm docstring).
  */
 
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import { TextSpliterator } from "spliterator"
 
-import {
-	alignAndWrite,
-	foldNOSurface,
-	makeMulberry32,
-	readTuples,
-	type CorpusRecipe,
-	sliceSourceID,
-} from "#recipes/scaffold"
+import { alignAndWrite, foldNOSurface, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 
 /**
  * Title-case a Kartverket ALL-CAPS locality (HELLVIK -> Hellvik); #690, all-caps is OOD.

package/lib/recipes/no-street-led.ts

@@ -16,16 +16,10 @@
  *   3. pc-first   "«pc» «city», «st» «n»"   (the 7% floor — the anchor the others converge to)
  */
 
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import { TextSpliterator } from "spliterator"
 
-import {
-	alignAndWrite,
-	foldNOSurface,
-	makeMulberry32,
-	readTuples,
-	type CorpusRecipe,
-	sliceSourceID,
-} from "#recipes/scaffold"
+import { alignAndWrite, foldNOSurface, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 
 /**
  * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,

package/lib/recipes/no-street.ts

@@ -10,9 +10,10 @@
  *   scripts/build-no-street-slice.mjs.
  */
 
+import { makeLcg } from "@mailwoman/core/utils"
+
 import {
 	alignAndWrite,
-	makeLcg,
 	readTuples,
 	sliceSourceID,
 	SYNTHETIC_TUPLE_LICENSE as LICENSE,

package/lib/recipes/po-box-cedex.ts

@@ -38,11 +38,11 @@ import { isPOBox } from "@mailwoman/codex/us"
 import { dataRootPath } from "@mailwoman/core/data-root"
 import { pathExists } from "@mailwoman/core/fs/readers"
 import { readZipEntry } from "@mailwoman/core/fs/zip"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 import { TextSpliterator, TSVSpliterator } from "spliterator"
 
 import {
-	makeMulberry32,
 	readCSVRecords,
 	readZippedCSVRecords,
 	sliceSourceID,

package/lib/recipes/po-box.ts

@@ -9,9 +9,10 @@
  *   scripts/build-po-box-slice.mjs.
  */
 
+import { makeLcg } from "@mailwoman/core/utils"
+
 import {
 	alignAndWrite,
-	makeLcg,
 	readTuples,
 	sliceSourceID,
 	SYNTHETIC_TUPLE_LICENSE as LICENSE,

package/lib/recipes/scaffold.ts

@@ -79,7 +79,6 @@ export interface SliceTuple {
  * A recipe must seed the same one its `.mjs` did, the same way it did (usually `seed`, but some derive a per-stream
  * seed), or `--seed N` stops being byte-reproducible.
  */
-export { makeLcg, mulberry32 as makeMulberry32, makeLcg as makeRandom } from "@mailwoman/core/utils"
 
 /**
  * One CSV record, keyed by its lower-cased header name, every value trimmed. A column the header does not declare reads

package/lib/recipes/si-bare-village.ts

@@ -24,7 +24,9 @@
  *   trailing «V» = locality (the binding the resolver needs).
  */
 
-import { alignAndWrite, makeMulberry32, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
+
+import { alignAndWrite, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 
 /**
  * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,

package/lib/recipes/street-affix.ts

@@ -35,14 +35,15 @@ import {
 	matchTrailingSuffix,
 	renderDirectional,
 	US_STREET_SUFFIX_PREFERRED_ABBR,
+	NAME_PRONE_US_SUFFIXES,
 } from "@mailwoman/codex/us"
 import { dataRootPath } from "@mailwoman/core/data-root"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 
 import { stableSourceID } from "#adapters/utils"
-import { NAME_PRONE_US_SUFFIXES } from "#name-prone-us-suffixes"
-import { makeMulberry32, readCSVRecords, readOATuples, sliceSourceID, type CorpusRecipe } from "#recipes/scaffold"
+import { readCSVRecords, readOATuples, sliceSourceID, type CorpusRecipe } from "#recipes/scaffold"
 import { pick } from "#synthesizers/utils"
 import type { CanonicalRow } from "#types"
 import { alignRow } from "#utils"

package/lib/recipes/street-bare.ts

@@ -17,7 +17,9 @@
  *   reproduces the legacy run byte-for-byte.
  */
 
-import { makeMulberry32, sliceSourceID, type CanonicalSliceRow, type CorpusRecipe } from "#recipes/scaffold"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
+
+import { sliceSourceID, type CanonicalSliceRow, type CorpusRecipe } from "#recipes/scaffold"
 import { DEFAULT_US_BASES } from "#synthesizers/intersection"
 import { synthesizeStreetRow, type StreetBaseTuple } from "#synthesizers/street"
 import { alignRow } from "#utils"

package/lib/recipes/street.ts

@@ -7,8 +7,10 @@
  *   {@link synthesizeStreetRow} → aligned LabeledRow. Ported from scripts/build-street-slice.mjs.
  */
 
+import { makeLcg } from "@mailwoman/core/utils"
+
 import { stableSourceID } from "#adapters/utils"
-import { alignAndWrite, makeLcg, readTuples, type CorpusRecipe } from "#recipes/scaffold"
+import { alignAndWrite, readTuples, type CorpusRecipe } from "#recipes/scaffold"
 import { synthesizeStreetRow, type StreetBaseTuple } from "#synthesizers/street"
 
 /**

package/lib/recipes/sub-venue-sources.ts

@@ -28,12 +28,13 @@
 import { dataRootPath } from "@mailwoman/core/data-root"
 import { readLocalJSONFile } from "@mailwoman/core/fs/readers"
 import { resolveModulePath } from "@mailwoman/core/module/resolvers"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { POIDatabase } from "@mailwoman/resolver-wof-sqlite/poi"
 import { DatabaseClient } from "@mailwoman/sqlite/client"
 import type { PathBuilderLike } from "path-ts"
 
 import { readTuples as readLocaleTuples, type LocalePart } from "#recipes/locale"
-import { makeMulberry32, readTuples as readSliceTuples } from "#recipes/scaffold"
+import { readTuples as readSliceTuples } from "#recipes/scaffold"
 import type { LocaleBaseTuple } from "#synthesizers/german"
 import { pick, weightedPick } from "#synthesizers/utils"
 import {

package/lib/recipes/sub-venue.ts

@@ -63,9 +63,10 @@
 import { dataRootPath } from "@mailwoman/core/data-root"
 import { isPresent } from "@mailwoman/core/objects"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 
-import { makeMulberry32, sliceSourceID, type CorpusRecipe } from "#recipes/scaffold"
+import { sliceSourceID, type CorpusRecipe } from "#recipes/scaffold"
 import {
 	buildIdentifierModel,
 	buildStreetNegatives,

package/lib/recipes/trailing-region.ts

@@ -66,14 +66,9 @@
  *   fact about a place; the postcode is not, because it does.
  */
 
-import {
-	alignAndWrite,
-	makeMulberry32,
-	type PostcodePlacement,
-	readTuples,
-	type CorpusRecipe,
-	sliceSourceID,
-} from "#recipes/scaffold"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
+
+import { alignAndWrite, type PostcodePlacement, readTuples, type CorpusRecipe, sliceSourceID } from "#recipes/scaffold"
 
 /**
  * Slice recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,

package/lib/recipes/unit.ts

@@ -24,14 +24,14 @@
 import { US_UNIT_DESIGNATOR_PREFERRED_ABBR, type USUnitDesignator } from "@mailwoman/codex/us"
 import { dataRootPath } from "@mailwoman/core/data-root"
 import type { ComponentTag } from "@mailwoman/core/types"
+import { mulberry32 as makeMulberry32 } from "@mailwoman/core/utils"
 import type { PathBuilderLike } from "path-ts"
 
 import { stableSourceID } from "#adapters/utils"
-import { makeMulberry32, readOATuples, type CorpusRecipe } from "#recipes/scaffold"
+import { readOATuples, type CorpusRecipe } from "#recipes/scaffold"
 import { pick } from "#synthesizers/utils"
 import type { CanonicalRow } from "#types"
 import { alignRow } from "#utils"
-
 /**
  * A cached OpenAddresses extract: the zip, the CSV member, and the implied (file-level) region.
  */

package/lib/tools/fetch/acra-sg.ts

@@ -0,0 +1,174 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   Fetch Singapore's ACRA corporate-entity register from data.gov.sg: 27 monthly CSVs (one per first
+ *   letter of the entity name plus "others"), 53 columns, the registered address FIELDED — `Block`,
+ *   `Street Name`, `Level No`, `Unit No`, `Building Name`, `Postal Code` — which is the NPPES shape,
+ *   so the corpus adapter renders the string and the spans fall out by construction. A Singapore
+ *   postcode names one building, so the postcode column is also the join to the OneMap/Overture rows.
+ *
+ *   License: Singapore Open Data Licence version 1.0 (attribution; commercial use, modification and
+ *   adaptation permitted; no share-alike). The prescribed attribution sentence goes in the manifest.
+ *
+ *   data.gov.sg serves a dataset in two calls: `initiate-download` prepares a signed URL and
+ *   `poll-download` answers it once the export is ready.
+ *
+ *   Invoke via `mailwoman corpus fetch acra-sg --out-root <path>`.
+ */
+
+import { BYTES_PER_KIB } from "@mailwoman/core/fs/formatters"
+import { pathExists } from "@mailwoman/core/fs/readers"
+import { makeDirectories } from "@mailwoman/core/fs/writers"
+import { sha256File } from "@mailwoman/core/hash"
+import { sleep } from "@mailwoman/core/utils/sleep"
+import { join } from "path-ts"
+
+import type { BaseFetchOptions, FetchSummary, SourceCollectionManifest, SourceManifest } from "#tools/fetch/download"
+import { loadCollectionFiles, streamBodyToFile, withRetries, writeManifest } from "#tools/fetch/download"
+
+const SLUG = "acra-sg"
+const COLLECTION_URL = "https://api-production.data.gov.sg/v2/public/api/collections/2/metadata"
+const DATASET_API = "https://api-production.data.gov.sg/v2/public/api/datasets"
+const DOWNLOAD_API = "https://api-open.data.gov.sg/v1/public/api/datasets"
+const LICENSE = "Singapore Open Data Licence version 1.0 — https://data.gov.sg/open-data-licence"
+
+const ATTRIBUTION =
+	"Contains information from ACRA Information on Corporate Entities accessed on <date> from data.gov.sg which is made available under the terms of the Singapore Open Data Licence version 1.0 https://data.gov.sg/open-data-licence"
+
+const POLL_INTERVAL_MS = 3000
+const POLL_ATTEMPTS = 40
+
+export type FetchACRASGOptions = BaseFetchOptions
+
+interface CollectionMetadata {
+	data?: { collectionMetadata?: { childDatasets?: string[] } }
+}
+
+interface DatasetMetadata {
+	data?: { name?: string; lastUpdatedAt?: string }
+}
+
+interface PollResponse {
+	data?: { status?: string; url?: string }
+}
+
+async function readJSON<T>(url: string): Promise<T> {
+	const res = await fetch(url, { headers: { accept: "application/json" } })
+
+	if (!res.ok) throw new Error(`acra-sg: ${url} answered HTTP ${res.status}`)
+
+	return (await res.json()) as T
+}
+
+/**
+ * Ask the portal to prepare the CSV export, then poll until it names the signed URL.
+ */
+async function signedURLFor(datasetID: string): Promise<string | undefined> {
+	await readJSON(`${DOWNLOAD_API}/${datasetID}/initiate-download`)
+
+	for (let attempt = 0; attempt < POLL_ATTEMPTS; attempt++) {
+		await sleep(POLL_INTERVAL_MS)
+		const poll = await readJSON<PollResponse>(`${DOWNLOAD_API}/${datasetID}/poll-download`)
+
+		if (poll.data?.status === "DOWNLOAD_SUCCESS" && poll.data.url) return poll.data.url
+	}
+
+	return undefined
+}
+
+function filenameFor(name: string, datasetID: string): string {
+	const letter = /\('([^']+)'\)/.exec(name)?.[1]
+
+	return `acra-entities-${(letter ?? datasetID).toLowerCase().replaceAll(/[^a-z0-9]+/g, "-")}.csv`
+}
+
+export async function fetchACRASG(options: FetchACRASGOptions, report?: (line: string) => void): Promise<FetchSummary> {
+	const destDir = join(options.outRoot, SLUG)
+	await makeDirectories(destDir)
+	const manifestPath = join(destDir, "MANIFEST.json")
+
+	const collection = await readJSON<CollectionMetadata>(COLLECTION_URL)
+	const datasetIDs = collection.data?.collectionMetadata?.childDatasets ?? []
+	report?.(`=== ${SLUG}: ${datasetIDs.length} datasets in collection 2`)
+
+	const previous = await loadCollectionFiles(manifestPath)
+	const files = new Map<string, SourceManifest>()
+	let fetched = 0
+	let skipped = 0
+	const failedCodes: string[] = []
+
+	for (const datasetID of datasetIDs) {
+		const metadata = await readJSON<DatasetMetadata>(`${DATASET_API}/${datasetID}/metadata`)
+		const name = metadata.data?.name ?? datasetID
+		const filename = filenameFor(name, datasetID)
+		const dest = join(destDir, filename)
+		const before = previous.get(filename)
+
+		if (before && (await pathExists(dest))) {
+			files.set(filename, before)
+
+			skipped++
+
+			continue
+		}
+
+		report?.(`--- ${name} (${datasetID})`)
+		let bytes: number
+
+		try {
+			bytes = await withRetries(
+				async () => {
+					const url = await signedURLFor(datasetID)
+
+					if (!url) throw new Error(`the export never reached DOWNLOAD_SUCCESS for ${datasetID}`)
+					const res = await fetch(url, { signal: AbortSignal.timeout(1_800_000) })
+
+					if (!res.ok) throw new Error(`HTTP ${res.status} on the signed URL for ${datasetID}`)
+
+					return streamBodyToFile(res, dest)
+				},
+				{ report, label: name }
+			)
+		} catch (error) {
+			report?.(`  ✗ ${(error as Error).message}`)
+			failedCodes.push(datasetID)
+
+			continue
+		}
+
+		if (bytes < BYTES_PER_KIB) {
+			report?.(`  ✗ ${bytes} bytes — an error page, not the register`)
+			failedCodes.push(datasetID)
+
+			continue
+		}
+
+		const sha = await sha256File(dest)
+
+		files.set(filename, {
+			source_url: `https://data.gov.sg/datasets/${datasetID}/view`,
+			downloaded_at: new Date().toISOString(),
+			filename,
+			sha256: sha,
+			bytes,
+		})
+
+		fetched++
+		report?.(`  ✓ ${(bytes / 1024 / 1024).toFixed(1)} MB  sha256=${sha}`)
+
+		const manifest: SourceCollectionManifest = {
+			source: SLUG,
+			source_url: "https://data.gov.sg/collections/2/view",
+			license: LICENSE,
+			attribution: ATTRIBUTION,
+			downloaded_at: new Date().toISOString(),
+			files: [...files.values()],
+		}
+
+		await writeManifest(manifestPath, manifest)
+	}
+
+	return { fetched, skipped, failed: failedCodes.length, failedCodes }
+}