@mailwoman/corpus 8.1.0 → 8.2.0

package/src/shard-recipes/intersection.ts

@@ -51,16 +51,20 @@ interface County {
 	state: string
 	regime: string
 }
+
 const TRAIN_COUNTIES: readonly County[] = [
 	{ fips: "17031", state: "IL", regime: "grid-city" },
 	{ fips: "34027", state: "NJ", regime: "suburb" },
 ]
+
 const GOLDEN_COUNTIES: readonly County[] = [{ fips: "50023", state: "VT", regime: "rural" }]
 
 const EVAL_GOLD_PATH = repoRootPath("data", "eval", "external", "intersection-real.jsonl")
 const OA_COOK = { zip: "/tmp/oa-cache/us__il__cook.zip", csv: "us/il/cook.csv" }
 
-/** One real crossing extracted from a county's TIGER EDGES shapefile. */
+/**
+ * One real crossing extracted from a county's TIGER EDGES shapefile.
+ */
 interface Crossing {
 	a: string
 	b: string
@@ -79,6 +83,7 @@ interface Form {
 	w: number
 	render: (a: string, b: string) => string
 }
+
 const FORMS: readonly Form[] = [
 	{ id: "amp", w: 0.2, render: (a, b) => `${a} & ${b}` },
 	{ id: "and", w: 0.2, render: (a, b) => `${a} and ${b}` },
@@ -100,6 +105,7 @@ interface Tail {
 	id: string
 	w: number
 }
+
 const TAILS: readonly Tail[] = [
 	{ id: "bare", w: 0.55 },
 	{ id: "region", w: 0.16 },
@@ -113,6 +119,7 @@ interface Casing {
 	w: number
 	apply: (s: string) => string
 }
+
 const CASES: readonly Casing[] = [
 	{ id: "as-is", w: 0.82, apply: (s) => s },
 	{ id: "upper", w: 0.12, apply: (s) => s.toUpperCase() },
@@ -131,7 +138,9 @@ const CONNECTOR_O_TOKENS = new Set(["and", "at", "of", "corner", "intersection"]
  */
 const BAD_NAME = /[,&@/]|\b(and|at)\b/i
 
-/** Punctuation a connector form may leave between spans (besides whitespace): `, & @ /`. */
+/**
+ * Punctuation a connector form may leave between spans (besides whitespace): `, & @ /`.
+ */
 const CONNECTOR_PUNCT_RE = /^[\s,&@/]*$/
 
 function weightedPick<T extends { w: number }>(items: readonly T[], random: () => number): T {
@@ -144,10 +153,12 @@ function weightedPick<T extends { w: number }>(items: readonly T[], random: () =
 		if (r <= 0) return item
 	}
 
-	return items[items.length - 1]!
+	return items.at(-1)!
 }
 
-/** Minimal RFC-4180-ish splitter (handles quoted fields) — same as the affix builder. */
+/**
+ * Minimal RFC-4180-ish splitter (handles quoted fields) — same as the affix builder.
+ */
 function splitCSV(line: string): string[] {
 	const out: string[] = []
 	let cur = ""
@@ -160,6 +171,7 @@ function splitCSV(line: string): string[] {
 			if (c === '"') {
 				if (line[i + 1] === '"') {
 					cur += '"'
+
 					i++
 				} else {
 					inQ = false
@@ -176,15 +188,20 @@ function splitCSV(line: string): string[] {
 			cur += c
 		}
 	}
+
 	out.push(cur)
 
 	return out
 }
 
-/** Order-insensitive crossing key, for eval-leakage exclusion + pair dedup. */
-const pairKey = (a: string, b: string): string => [a.toLowerCase(), b.toLowerCase()].sort().join("\x1f")
+/**
+ * Order-insensitive crossing key, for eval-leakage exclusion + pair dedup.
+ */
+const pairKey = (a: string, b: string): string => [a.toLowerCase(), b.toLowerCase()].toSorted().join("\u001F")
 
-/** Load the eval's crossings so neither train nor golden ever sees them. */
+/**
+ * Load the eval's crossings so neither train nor golden ever sees them.
+ */
 function readEvalExclusions(): { nodes: Set<number>; pairs: Set<string> } {
 	const nodes = new Set<number>()
 	const pairs = new Set<string>()
@@ -217,6 +234,7 @@ async function extractCrossings(
 	seed: number
 ): Promise<Crossing[]> {
 	const shp = `${edgesDir}/tl_2023_${county.fips}_edges.shp`
+
 	const result = await db.runAndReadAll(`
 		WITH incidence AS (
 			SELECT TNIDF AS node, FULLNAME AS name, ZIPL AS zip
@@ -238,6 +256,7 @@ async function extractCrossings(
 		WHERE len(names[1]) >= 6 AND len(names[2]) >= 6
 		ORDER BY h
 	`)
+
 	const out: Crossing[] = []
 
 	for (const r of result.getRowObjects()) {
@@ -254,7 +273,9 @@ async function extractCrossings(
 	return out
 }
 
-/** ZIP → majority city from the cached OA Cook-county CSV (real ZIP/city pairings). */
+/**
+ * ZIP → majority city from the cached OA Cook-county CSV (real ZIP/city pairings).
+ */
 function buildZipCityMap(): Map<string, string> {
 	const r = spawnSync("unzip", ["-p", OA_COOK.zip, OA_COOK.csv], { maxBuffer: 1024 * 1024 * 1024, encoding: "buffer" })
 
@@ -263,6 +284,7 @@ function buildZipCityMap(): Map<string, string> {
 
 		return new Map()
 	}
+
 	const lines = r.stdout.toString("utf8").split(/\r?\n/)
 
 	if (lines.length < 2) return new Map()
@@ -284,8 +306,10 @@ function buildZipCityMap(): Map<string, string> {
 		if (!byCity) {
 			counts.set(zip, (byCity = new Map()))
 		}
+
 		byCity.set(city, (byCity.get(city) ?? 0) + 1)
 	}
+
 	const map = new Map<string, string>()
 
 	for (const [zip, byCity] of counts) {
@@ -392,7 +416,7 @@ function auditRow(row: LabeledRow, components: Partial<Record<ComponentTag, stri
 		}
 	}
 
-	if (errors.length > 0) return errors
+	if (errors.length) return errors
 
 	const compCount = Object.keys(components).length
 
@@ -407,8 +431,10 @@ function auditRow(row: LabeledRow, components: Partial<Record<ComponentTag, stri
 
 		if (indices.length !== 1) {
 			errors.push(`${tag}: expected 1 span, got ${indices.length}`)
+
 			continue
 		}
+
 		const idx = indices[0]!
 		const got = raw.slice(span_starts[idx]!, span_ends[idx]!)
 
@@ -425,6 +451,7 @@ function auditRow(row: LabeledRow, components: Partial<Record<ComponentTag, stri
 		if (span_starts[i]! > cursor) {
 			uncovered.push(raw.slice(cursor, span_starts[i]!))
 		}
+
 		cursor = span_ends[i]!
 	}
 
@@ -440,7 +467,8 @@ function auditRow(row: LabeledRow, components: Partial<Record<ComponentTag, stri
 				errors.push(`illegal uncovered word "${word}"`)
 			}
 		}
-		const punctOnly = segment.replace(/[\p{L}\p{N}]+/gu, "")
+
+		const punctOnly = segment.replaceAll(/[\p{L}\p{N}]+/gu, "")
 
 		if (!CONNECTOR_PUNCT_RE.test(punctOnly)) {
 			errors.push(`illegal uncovered punctuation in "${segment}"`)
@@ -450,6 +478,10 @@ function auditRow(row: LabeledRow, components: Partial<Record<ComponentTag, stri
 	return errors
 }
 
+/**
+ * Shard recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
 export const intersectionRecipe: ShardRecipe = {
 	name: "intersection",
 	description: "Real-pair intersection rows (US): TIGER 2023 EDGES crossings → audited intersection_a/b labels",
@@ -458,11 +490,12 @@ export const intersectionRecipe: ShardRecipe = {
 	async run(opts, write) {
 		// Legacy build-intersection-shard.mjs seeded `mulberry32(opts.seed)`.
 		const random = makeMulberry32(opts.seed)
-		const count = opts.count ?? 40000
+		const count = opts.count ?? 40_000
 		const source = opts.sourceName ?? "synth-intersection"
 		const edgesDir = opts.edgesDir ?? "/tmp/tiger-edges"
 		const counties = opts.golden ? GOLDEN_COUNTIES : TRAIN_COUNTIES
 		const exclusions = readEvalExclusions()
+
 		console.error(`  eval exclusions: ${exclusions.nodes.size} nodes, ${exclusions.pairs.size} pairs`)
 
 		const { DuckDBInstance } = await import("@duckdb/node-api")
@@ -484,26 +517,32 @@ export const intersectionRecipe: ShardRecipe = {
 
 				if (exclusions.nodes.has(c.node) || exclusions.pairs.has(key)) {
 					stats.evalExcluded++
+
 					continue
 				}
 
 				if (BAD_NAME.test(c.a) || BAD_NAME.test(c.b) || c.a.includes(c.b) || c.b.includes(c.a)) {
 					stats.badName++
+
 					continue
 				}
 
 				if (seenPairs.has(key)) {
 					stats.dupPair++
+
 					continue
 				}
+
 				seenPairs.add(key)
 				pool.push(c)
+
 				kept++
 			}
+
 			console.error(`  ${county.fips} (${county.state}, ${county.regime}): ${crossings.length} crossings, ${kept} kept`)
 		}
 
-		if (pool.length === 0) {
+		if (!pool.length) {
 			throw new Error(`No crossings found — are the TIGER EDGES shapefiles present in ${edgesDir}?`)
 		}
 
@@ -531,6 +570,7 @@ export const intersectionRecipe: ShardRecipe = {
 
 			if (seenRaw.has(raw)) {
 				skipped++
+
 				continue
 			}
 
@@ -542,7 +582,9 @@ export const intersectionRecipe: ShardRecipe = {
 				caseCounts[caseID] = (caseCounts[caseID] ?? 0) + 1
 				countyCounts[crossing.fips] = (countyCounts[crossing.fips] ?? 0) + 1
 				usedCrossings.add(crossing.node)
+
 				emitted++
+
 				continue
 			}
 
@@ -556,18 +598,22 @@ export const intersectionRecipe: ShardRecipe = {
 				corpus_version: "0.4.0",
 				license: "TIGER/Line 2023 EDGES (US Census, public domain) real street pairs; OA Cook IL zip-to-city tails",
 			}
+
 			// Verbatim-only alignment: raw is built from the component values, so a fuzzy fallback could
 			// only ever mislabel (e.g. claim a lookalike window for a near-duplicate street).
 			const aligned = alignRow(canonical, { maxEditDistance: 0 })
 
 			if (aligned.kind !== "labeled" || !aligned.row) {
 				skipped++
+
 				continue
 			}
+
 			const violations = auditRow(aligned.row, components)
 
-			if (violations.length > 0) {
+			if (violations.length) {
 				auditErrors.push({ raw, violations })
+
 				continue
 			}
 
@@ -582,6 +628,7 @@ export const intersectionRecipe: ShardRecipe = {
 			if (samples.length < FORMS.length && !samples.some((s) => s.form === formID)) {
 				samples.push({ form: formID, raw, tokens: aligned.row.tokens, labels: aligned.row.labels })
 			}
+
 			emitted++
 		}
 
@@ -599,7 +646,9 @@ export const intersectionRecipe: ShardRecipe = {
 			source: "TIGER2023 EDGES via DuckDB ST_Read; node = 2 distinct S1* FULLNAMEs; eval crossings excluded",
 			samples,
 		}
+
 		writeFileSync(opts.output.replace(/\.jsonl$/, ".report.json"), JSON.stringify(report, null, "\t"))
+
 		console.error(
 			`Done: emitted ${emitted} rows (skipped ${skipped}) from ${usedCrossings.size}/${pool.length} real crossings. → ${opts.output}\n` +
 				`  forms: ${JSON.stringify(formCounts)}\n` +
@@ -608,7 +657,7 @@ export const intersectionRecipe: ShardRecipe = {
 				`  audit: ${auditErrors.length} violation(s)`
 		)
 
-		if (auditErrors.length > 0) {
+		if (auditErrors.length) {
 			throw new Error(`AUDIT FAILED — first violation: ${JSON.stringify(auditErrors[0])}`)
 		}
 

package/src/shard-recipes/locale.ts

@@ -239,11 +239,14 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 		input = createReadStream(part.path)
 	} else {
 		const child = spawn("unzip", ["-p", part.zip!, part.csv!])
+
 		child.on("error", (err) => {
 			console.error(`  WARN: unzip failed for ${part.zip}: ${err.message}`)
 		})
+
 		input = child.stdout!
 	}
+
 	const get = (cells: string[], i: number): string => (i >= 0 && i < cells.length ? (cells[i] ?? "").trim() : "")
 	const reservoir: LocaleBaseTuple[] = []
 	let cols: ColumnIndex | null = null
@@ -263,7 +266,9 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 		})) {
 			if (header === null) {
 				header = cells.map((h) => h.trim().toLowerCase())
+				// oxlint-disable-next-line no-loop-func -- the binding is per-iteration (for-of/for-await) and the batch is awaited before the next
 				const ix = (name: string): number => header!.indexOf(name)
+
 				// `cnigRaw` (ES pedanía only): the RAW CNIG header has no NUMBER/STREET/CITY/DISTRICT/REGION/POSTCODE
 				// at all — `numero`/`nombre_via`/`poblacion`/`municipio`/`comunidad_autonoma`/`cod_postal` instead.
 				// See {@link LocalePart.cnigRaw}.
@@ -291,6 +296,7 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 			}
 
 			if (cols === null) continue
+
 			// cnigRaw: STREET is split into a road-type column (`tipo_vial`, e.g. "CARRETERA") and the name
 			// (`nombre_via`) — rejoin them exactly as OA's own conform step does (verified byte-identical to the
 			// conformed STREET column for the same source row). Every other part's header already carries the
@@ -299,6 +305,7 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 				cols.tipoVial >= 0
 					? [get(cells, cols.tipoVial), get(cells, cols.street)].filter(Boolean).join(" ")
 					: get(cells, cols.street)
+
 			const rawCity = get(cells, cols.city)
 
 			if (!street) continue
@@ -324,6 +331,7 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 				if (cleanedDistrict) {
 					locality = cleanedDistrict
 					const cleanedCity = cleanCityNoise(rawCity)
+
 					// ES pedanía lesson (2026-07-22): the CNIG `poblacion` column is filled on ~93% of rows but
 					// EQUALS `municipio` on the majority of those (the address point sits in the municipio's own
 					// main town, not a below-municipio pedanía) — only ~32.6% of ES rows carry a genuinely
@@ -348,6 +356,7 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 
 				continue
 			}
+
 			const tuple: LocaleBaseTuple = {
 				house_number: get(cells, cols.num),
 				street,
@@ -356,6 +365,7 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 				postcode: get(cells, cols.post),
 				...(dependent_locality ? { dependent_locality } : {}),
 			}
+
 			seen++
 
 			if (reservoir.length < RESERVOIR_CAP) {
@@ -369,8 +379,8 @@ export async function readTuples(part: LocalePart, rng: () => number): Promise<L
 				}
 			}
 		}
-	} catch (err) {
-		console.error(`  WARN: read failed for ${part.path ?? part.zip}: ${(err as Error).message}`)
+	} catch (error) {
+		console.error(`  WARN: read failed for ${part.path ?? part.zip}: ${(error as Error).message}`)
 
 		return []
 	}
@@ -429,6 +439,10 @@ export function resolveLocaleParts(countrySource: LocaleCountrySource, override:
 	return override === true && countrySource.pedaniaParts ? countrySource.pedaniaParts : countrySource.parts
 }
 
+/**
+ * Shard recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
 export const localeRecipe: ShardRecipe = {
 	name: "locale",
 	description: "Per-locale coverage rows (DE/FR/NL/IT/ES/NZ/GB) from real OA tuples, both orders → synthesizeLocaleRow",
@@ -459,11 +473,13 @@ export const localeRecipe: ShardRecipe = {
 				`No OA sources registered for --country ${country}. Known: ${Object.keys(COUNTRY_SOURCES).join(", ")}.`
 			)
 		}
+
 		const intlFraction = opts.intlFraction ?? 0.4
 
 		if (!(intlFraction >= 0 && intlFraction <= 1)) {
 			throw new Error(`--intl-fraction must be in [0, 1], got ${intlFraction}`)
 		}
+
 		// Default 0 → the `random() < countryFraction` draw below is short-circuited away entirely (never
 		// consumed), so every existing locale's emit stream is byte-identical to before this option existed.
 		const countryFraction = opts.countryFraction ?? 0
@@ -471,6 +487,7 @@ export const localeRecipe: ShardRecipe = {
 		if (!(countryFraction >= 0 && countryFraction <= 1)) {
 			throw new Error(`--country-fraction must be in [0, 1], got ${countryFraction}`)
 		}
+
 		const source = opts.sourceName ?? countrySource.source
 		const count = opts.count ?? 4000
 		// Tri-state: `undefined` (flag absent) touches nothing below — `parts` stays the default list and each
@@ -485,7 +502,7 @@ export const localeRecipe: ShardRecipe = {
 		for (let pi = 0; pi < parts.length; pi++) {
 			// A reservoir PRNG per part, seeded but independent of the emit loop's `random`, so the sample is
 			// reproducible without perturbing the synth/order draws.
-			const reservoirRng = makeMulberry32((opts.seed ^ (0x9e3779b9 * (pi + 1))) >>> 0)
+			const reservoirRng = makeMulberry32((opts.seed ^ (0x9e_37_79_b9 * (pi + 1))) >>> 0)
 			const effectivePart = applyDistrictAsLocalityOverride(parts[pi]!, districtAsLocalityOverride)
 			const t = await readTuples(effectivePart, reservoirRng)
 
@@ -494,7 +511,7 @@ export const localeRecipe: ShardRecipe = {
 			} // NOT pool.push(...t) — spreading huge arrays overflows the stack
 		}
 
-		if (pool.length === 0) {
+		if (!pool.length) {
 			throw new Error(`No ${country} tuples found — are the source CSVs/zips present? (see COUNTRY_SOURCES)`)
 		}
 
@@ -506,22 +523,27 @@ export const localeRecipe: ShardRecipe = {
 		while (emitted < count && guard++ < count * 6) {
 			const base = pool[Math.floor(random() * N)]!
 			const order = random() < intlFraction ? "international" : "native"
+
 			// Per-country surface-shape draws (#241) — consumed ONLY for that country, so the DE/FR emit
 			// streams for a given seed are unchanged by their existence.
 			const nativeHouseJoin =
 				country === "ES" ? (random() < ES_SPACE_JOIN_FRACTION ? ("space" as const) : ("template" as const)) : undefined
+
 			const postcodeShape =
 				country === "NL"
 					? random() < NL_GLUED_POSTCODE_FRACTION
 						? ("as-source" as const)
 						: ("conventional" as const)
 					: undefined
+
 			const synth = synthesizeLocaleRow(base, country, { random, order, nativeHouseJoin, postcodeShape })
 
 			if (!synth) {
 				skipped++
+
 				continue
 			}
+
 			applyCountryAppend(synth, country, countryFraction, random)
 
 			if (opts.golden) {
@@ -535,22 +557,29 @@ export const localeRecipe: ShardRecipe = {
 					source,
 					source_id: "golden:align-check",
 				}
+
 				const goldenAligned = alignRow(goldenCanonical as Parameters<typeof alignRow>[0])
 
 				if (goldenAligned.kind !== "labeled" || !goldenAligned.row) {
 					skipped++
+
 					continue
 				}
+
 				write(JSON.stringify({ raw: synth.raw, components: synth.components, country, order }) + "\n")
+
 				emitted++
+
 				continue
 			}
+
 			const sourceID = stableSourceID(source, {
 				street: synth.components.street,
 				house_number: synth.components.house_number,
 				locality: synth.components.locality,
 				postcode: synth.components.postcode,
 			})
+
 			const canonical = {
 				raw: synth.raw,
 				components: synth.components,
@@ -561,13 +590,17 @@ export const localeRecipe: ShardRecipe = {
 				corpus_version: countrySource.corpusVersion,
 				license: `OpenAddresses ${country} tuples, rendered ${order}-order — see ingest SOURCES`,
 			}
+
 			const aligned = alignRow(canonical as Parameters<typeof alignRow>[0])
 
 			if (aligned.kind !== "labeled" || !aligned.row) {
 				skipped++
+
 				continue
 			}
+
 			write(JSON.stringify({ ...aligned.row, synth_method: source, synth_order: order, synth_base_id: null }) + "\n")
+
 			emitted++
 		}
 

package/src/shard-recipes/nl-postcode.ts

@@ -20,11 +20,17 @@
 
 import { alignAndWrite, makeMulberry32, readTuples, type ShardRecipe, shardSourceID } from "./scaffold.ts"
 
-/** "1012LG" → "1012 LG". The tuples carry the unspaced OA form; the spaced form is the failing case. */
+/**
+ * "1012LG" → "1012 LG". The tuples carry the unspaced OA form; the spaced form is the failing case.
+ */
 function spacePostcode(pc: string): string {
 	return pc.replace(/^(\d{4})([A-Z]{2})$/, "$1 $2")
 }
 
+/**
+ * Shard recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
 export const nlPostcodeRecipe: ShardRecipe = {
 	name: "nl-postcode",
 	description:
@@ -41,13 +47,15 @@ export const nlPostcodeRecipe: ShardRecipe = {
 			const street = String(t.street ?? "").trim()
 			const city = String(t.locality ?? "").trim()
 			const number = String(t.number ?? "").trim()
+
 			const rawPostcode = String(t.postcode ?? "")
 				.trim()
 				.toUpperCase()
-				.replace(/\s+/g, "")
+				.replaceAll(/\s+/g, "")
 
 			if (!street || !city || !number || !/^\d{4}[A-Z]{2}$/.test(rawPostcode)) {
 				skipped++
+
 				continue
 			}
 
@@ -75,12 +83,14 @@ export const nlPostcodeRecipe: ShardRecipe = {
 				postcode,
 				locality: city,
 			}
+
 			const source_id = shardSourceID("synth-nl-postcode", {
 				...components,
 				o: String(order),
 				s: spaced ? "1" : "0",
 				v: String(read),
 			})
+
 			const canonical = {
 				raw,
 				components,

package/src/shard-recipes/no-fragment.ts

@@ -44,15 +44,21 @@ import { alignAndWrite, makeMulberry32, readTuples, type ShardRecipe, shardSourc
  * collapse whitespace, KEEP diacritics. Stripping them (fr-fragment's norm) would fold `Tømmerlien` -> `tommerlien`,
  * miss the board's reserved `tømmerlien`, and leak the surface.
  */
-const norm = (value: string): string => value.normalize("NFC").toLowerCase().replace(/\s+/g, " ").trim()
+const norm = (value: string): string => value.normalize("NFC").toLowerCase().replaceAll(/\s+/g, " ").trim()
 
-/** Title-case a Kartverket ALL-CAPS locality (HELLVIK -> Hellvik); #690, all-caps is OOD. */
+/**
+ * Title-case a Kartverket ALL-CAPS locality (HELLVIK -> Hellvik); #690, all-caps is OOD.
+ */
 const titleNO = (value: string): string =>
 	value
 		.split(/\s+/)
 		.map((w) => (w ? w[0]!.toUpperCase() + w.slice(1).toLowerCase() : w))
 		.join(" ")
 
+/**
+ * Shard recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
 export const noFragmentRecipe: ShardRecipe = {
 	name: "no-fragment",
 	description:
@@ -123,6 +129,7 @@ export const noFragmentRecipe: ShardRecipe = {
 			// emitSeq keeps every emit distinct — knob 3 emits N copies of one long-number row, and
 			// (components, read) alone would collide their source_id and let downstream dedup drop the boost.
 			const source_id = shardSourceID("synth-no-fragment", { ...components, k: klass, v: `${read}:${emitSeq++}` })
+
 			const canonical = {
 				raw,
 				components,
@@ -158,12 +165,14 @@ export const noFragmentRecipe: ShardRecipe = {
 
 			if (!street) {
 				skipped++
+
 				continue
 			}
 
 			// THE SPLIT. A surface on the digit board never enters training.
 			if (excluded.has(norm(street))) {
 				contaminated++
+
 				continue
 			}
 
@@ -180,6 +189,7 @@ export const noFragmentRecipe: ShardRecipe = {
 
 					emit(pc, { postcode: pc }, "counter-bare-postcode")
 				}
+
 				continue
 			}
 

package/src/shard-recipes/no-street-led.ts

@@ -27,8 +27,12 @@ import { alignAndWrite, makeMulberry32, readTuples, type ShardRecipe, shardSourc
  * check would never match the board's reserved `tømmerlien` and the split would leak silently. Diacritic street heads
  * (…vegen/…veien with ø/å/æ) are the whole point of this shard's boundary; folding them away is not an option.
  */
-const norm = (value: string): string => value.normalize("NFC").toLowerCase().replace(/\s+/g, " ").trim()
+const norm = (value: string): string => value.normalize("NFC").toLowerCase().replaceAll(/\s+/g, " ").trim()
 
+/**
+ * Shard recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
 export const noStreetLedRecipe: ShardRecipe = {
 	name: "no-street-led",
 	description: "NO street-led boundary form (#901 family): '«st» «n», «pc» «city»' — diacritic street heads",
@@ -81,16 +85,20 @@ export const noStreetLedRecipe: ShardRecipe = {
 
 			if (!street || !city || !number || !postcode) {
 				skipped++
+
 				continue
 			}
 
 			// A surface on the digit board never enters training.
 			if (excluded.has(norm(street))) {
 				contaminated++
+
 				continue
 			}
+
 			const order = read % 3
 			let raw: string
+
 			const components: Record<string, string> = {
 				street,
 				house_number: number,
@@ -105,7 +113,9 @@ export const noStreetLedRecipe: ShardRecipe = {
 			} else {
 				raw = `${postcode} ${city}, ${street} ${number}`
 			}
+
 			const source_id = shardSourceID("synth-no-street-led", { ...components, o: String(order), v: String(read) })
+
 			const canonical = {
 				raw,
 				components,

package/src/shard-recipes/no-street.ts

@@ -15,6 +15,10 @@ import { alignAndWrite, makeLcg, readTuples, shardSourceID, type ShardRecipe } f
 
 const LICENSE = "Synthetic — derived from CC-BY / public-domain input tuples"
 
+/**
+ * Shard recipe registered with the corpus builder — see the file header for the parse behaviour it exists to exercise,
+ * and `description` below for the surface form it generates.
+ */
 export const noStreetRecipe: ShardRecipe = {
 	name: "no-street",
 	description: "No-street counter-example rows: tuples → synthesizeNoStreetRow → aligned LabeledRow",
@@ -33,6 +37,7 @@ export const noStreetRecipe: ShardRecipe = {
 
 			if (!tuple.locality || !tuple.region || !tuple.postcode || !tuple.country) {
 				skipped++
+
 				continue
 			}
 
@@ -41,8 +46,10 @@ export const noStreetRecipe: ShardRecipe = {
 
 				if (!synth) {
 					skipped++
+
 					continue
 				}
+
 				const ok = alignAndWrite(
 					write,
 					{