Suriname (SR) — internal licensing notes

Status: BUILT LOCALLY, free-host, NOT PUBLISHED — 2026-08-27. Built per instruction "ok build suriname, free host, validate and ship, only local". Nothing pushed to git or R2.

Product identity

Suriname has no postal code system — confirmed against the project's cached Wikipedia postal-countries list (SUR -> has_postal_codes: false) and by https://download.geonames.org/export/zip/SR.zip returning HTTP 404.

Ships as an admin-divisions-only product in the 13-column postal-codes schema, postal_code empty on every row — the same deliberate exception to the product definition already agreed for SC and CD. Not the FedEx-side 00000 convention: the column is blank, not placeholdered.

Source and licence

• Source: GeoNames gazetteer dump for Suriname (https://download.geonames.org/export/dump/SR.zip), plus the GeoNames admin1CodesASCII.txt table.
• Licence: CC BY 4.0 — attribution only, no share-alike. licence_status.decision: "compatible".

This differs from its peers in a way that matters

CD and SC are free-only by constraint — both were built from Wikipedia (CC BY-SA), whose share-alike clause makes an exclusive sale impossible.

Suriname is free-only by policy. CC BY 4.0 would permit sale on attribution alone; it ships free because of the standing decision to free-host the no-postal-code countries. Do not read SR's free status as evidence its licence forbids sale — it does not. If that policy is ever revisited, SR is on the sellable side of the line and SC/CD are not.

Rejected source — do not substitute it

HDX / OCHA COD-AB "Suriname - Subnational Administrative Boundaries" is labelled cc-by-igo, which reads as commercially reusable. Its own methodology field says the data was "downloaded from www.gadb.org" — a typo for gadm.org. GADM is non-commercial. A redistributor's package label does not cure a non-commercial component.

This is the third COD-AB package found with this defect (Albania, Trinidad & Tobago, Suriname). Checked and CLEAN, for the record: Sierra Leone ("collected from the government and cleaned and pcoded by OCHA and ITOS" — and we sell SL, so this mattered), Guyana (DEVInfo LAC), Seychelles (Seychelles National Bureau of Statistics 2010 census), DR Congo (MONUC/UNDP). The taint is per-package and must be checked from methodology every time, never assumed from the licence label.

What was built

• sr.csv — 68 rows, one per ressort, all 10 districts present.
• admin_code1 is the official ISO 3166-2:SR code (BR CM CR MA NI PR PM SA SI WA), derived from the district name. GeoNames' own numeric district code (10–19) is internal and never published.
• admin_code2 is deliberately EMPTY. GeoNames' admin2 value for Suriname is its own geonameId (SR.10.12088671). Publishing an internal record number in a field consumers read as an official code is exactly the live defect in the published Albania file. An empty column is correct.
• Coordinates on 68/68 rows, 5 dp, accuracy: RESSORT — the ressort's own gazetteer point, labelled so it cannot be mistaken for street level. Better than CD and SC, which ship no coordinates at all.
• 14 rows carry alternate spellings; aliases echoing place_name are dropped.

Disclosed, not hidden — the row count exceeds the official one

Published counts of Suriname's ressorten are 62 or 63 (sources disagree with each other). GeoNames enumerates 68. This is a surplus, not a gap, and it could not be reconciled — no official Surinamese ressort register was reachable in machine-readable form. The README says this plainly and tells consumers to treat the file as a gazetteer of second-level units rather than a certified register. All 10 districts are present and none is under-represented.

Validation

• Build aborts if GeoNames' district set disagrees with the ISO table, if any ADM2 feature carries an unknown district, or on any duplicate (district, ressort) pair.
• Frictionless valid from inside data/, and valid again from the unzipped sr.zip as a customer receives it: declared rows 68 = actual 68, declared bytes 5,191 = actual, sha256 matches.

Published — 2026-08-27

• Repo: commit 86b46fd on datopian/postal-codes main (yml, notes, 4 scripts).
• postal-codes bucket: full package — sr.csv, datapackage.json, README.md, ATTRIBUTION.txt, sr.zip. All five sha256-match the local build.
• postal-codes-samples bucket: sr.csv only, and it is the FULL 68-row file, not a 100-row sample — SR is free-hosted, so there is no paid version to hold back. Live at https://postal.datahub.io/sr/sr.csv (HTTP 200, 5,191 bytes).
• Both _meta/catalog.json files updated, 248 entries each, previous versions backed up.

Samples bucket convention — got this wrong first time. The initial publish pushed all five files there. The bucket is CSV-only (257 of 274 objects are .csv, one per country); the only .txt and .zip in the whole bucket were the ones that publish added. Corrected by deleting the four extras and rewriting the SR catalog entry; publish_r2.py now carries a per-target files list so it cannot recur.

Open point for a human: ATTRIBUTION.txt therefore does not travel with the CSV on the public endpoint, while CC BY 4.0 makes attribution a condition of use. Attribution is present in the postal-codes package, in datapackage.yml, and on the dataset page; whether that satisfies the licence for the bare public CSV is a call worth making deliberately rather than by default — and it applies to all 257 CSVs in that bucket, not just SR.

Previously not done sr/ in the bucket still holds the

159-byte header-only stub from 2026-07-30, and datasets/sr/datapackage.yml has been updated locally (data_available: true, real sources[], licenses[]) but not committed. is_premium is correctly absent and must stay absent.