Internal record of the P6 migration from the old scripts to the producer: the discrepancies found, their causes, and who decided what. This is a dated record, so the numbers are as of the migration and not kept current. It is not shipped in the zip.
Issue: pc-28e.8.26
Commit: see git log -- datasets/zw/ (P6 zw commit on feat/single-source-of-truth, base e797eb1)
Parity verdict:format-only (gate: pass, exit 0); CRLF→LF only
Used only to check GeoNames' province labels against the crosswalk. The old builder read a local sources/geonames_admin1CodesASCII.txt snapshot and joined on it by name
Both URLs are stable, unversioned GeoNames URLs. The fetch date and sha256 are recorded in zw.inputs.json; --no-fetch reuses the cache.
Zimbabwe has no postal-code system and GeoNames has no postal export for ZW (export/zip/ZW.zip is a 404), so postal_code is empty on every row. pipeline.postal_code_required: false tells build.py this is intended (bj precedent).
Parity against the live public CSV (2026-10-05)
Rows: 2,101 live and 2,101 fresh, none only on one side, 0 changed cells, no cell classes, 0 duplicate keys. Once CRLF is normalised to LF the fresh zw.csv is byte-identical to the live CSV, including row order. Live sha256 c7f8001c…, fresh 544a432d… (the only difference is line endings).
No discrepancies to explain. The GeoNames dump at migration time still has 2,101 populated places, 2 of them on the reserved admin1 code 00 (Nyamapande, Sango), as on 2026-09-15.
accuracy is empty on every row (the gazetteer dump has no accuracy field), so it isn't numeric data and doesn't get the GeoNames accuracy description
primaryKey: [place_name, latitude, longitude]
Agent, from the produced CSV
Unique in the produced CSV (bj precedent). The old builder's PK also included admin_name1, which is redundant and blank on the 00 rows. (place_name, admin_code1) and (latitude, longitude) are not unique
pipeline.postal_code_required: false
Brief rule (bj precedent)
postal_code is empty on every row
Add admin1CodesASCII.txt to the descriptor sources
Agent
It is a fetched input
Field descriptions taken from the old builder's datapackage.json (minus counts)
Agent (bj precedent)
The P3 generic descriptions said "Postal code for the location" etc. The postal_code description no longer points at ATTRIBUTION.txt, which a free dataset doesn't ship
Remove the row count from the descriptor description and README
Brief doc rule
Province count ("all 10") kept
Transforms carried over
From build_postal_codes_product.py:
Feature class P only; place_name is the dump's name column; latitude/longitude copied verbatim.
admin_code1 → ISO 3166-2:ZW for the 10 provinces (verified against en.wikipedia.org/wiki/ISO_3166-2:ZW on 2026-09-15), admin_name1 the ISO name, which drops GeoNames' " Province" suffix (Midlands, Mashonaland East, Matabeleland South, Masvingo).
postal_code, accuracy, alternative_city_name and admin tiers 2–3 empty.
Sorted by (admin_code1, place_name), ties in dump order.
Not ported: the old builder's own datapackage.json, README, ATTRIBUTION and zip writing (build.py / committed docs handle those) and publish_r2.py.
Open questions
README.md was not the stale stub (it already describes the locality directory and says free), so only the row count was removed; no pending README rewrite.
README.md still says "See ATTRIBUTION.txt in the packaged data", but there is no ATTRIBUTION.txt in the repo for zw and free datasets ship no zip. Left alone; belongs to the pc-hfz sweep.
.licensing-notes.md still carries dated counts and the stale data_available: false / licence_status wording; left for pc-hfz.
API access
Access dataset files directly from scripts, code, or AI agents.
Each file has a stable URL (r-link) that you can use directly in scripts, apps, or AI agents. These URLs are permanent and safe to hardcode.