Published

Netherlands (NL) build pipeline

DataHub
DataHub
0

Postal Codes Dataset for Netherlands, NL including name of the city, town, or place, various administrative divisions and alternative city names.

Netherlands (NL) build pipeline

Not a live external fetch — converts an existing FedEx-schema delivery (~/Desktop/Research/fedex-agents-prototype/datahub-sales-deliveries/Netherlands_Delivery/Netherlands.csv, 467,630-row Universal Data Schema export, already QA'd upstream: validation_report.json there shows 0 errors) into this repo's 13-column postal-codes schema. convert_from_fedex.py's own docstring documents every place the two schemas disagree in detail — read it before changing anything. Summary:

  1. countryCode (NLD, alpha-3) -> country_code (NL, alpha-2).
  2. The 12-column FedEx schema has additionalInfo; the 13-column one has nowhere to put it. Only 10 of 467,630 rows use it — 8 say "Postal station: Cortelande" (genuinely lost in conversion, noted in data/README.md), 2 carry a licensing statement (see point 3).
  3. Licensing: the FedEx base source is CC0 1.0 (public domain, no attribution needed). Two rows were backfilled from the CBS woonplaatsen reference (CC-BY 4.0) — those two rows alone would drag the whole published file into an attribution obligation. They are exactly the two rows asserting postalCode "00000" (no postal code) — which don't belong in this product anyway (a list of postal codes has no home for a no-postal-code row; see CLAUDE.md, "Two Output Formats" in fedex-agents-prototype). Dropping them is correct on product grounds and keeps the published dataset cleanly CC0.
  4. place_name and admin_name3 both take cityName — legitimately the same value in NL (woonplaats is both the locality and the third admin tier). The pre-existing R2 nl.csv had admin_name3 as the literal string 'nan'; this replaces it with the real value.

Run order

python3 convert_from_fedex.py  # reads the FedEx delivery path above directly
python3 archive_nl_full.py     # archives the untruncated conversion before any drops
python3 integrity_checker.py
python3 package.py
python3 publish_r2.py          # dry run by default; --apply to publish

Verification (2026-08-19)

Confirmed the FedEx source file is real (not a stub) and ran the conversion fresh against a fresh copy of it: 467,628 rows (the 2 00000 rows correctly dropped per point 3 above), byte-identical (MD5 match) to both the pre-existing local copy and live R2. Publish dry-run correctly ignores the phantom nl-full.csv that _meta/catalog.json still references — that catalog staleness is isolated to the index, not this data or this script (see the separately-flagged stale-catalog issue for SG/DE/PT/NL/IT).