openg2p-data

Sample and reference data for the OpenG2P platform, plus the generators that produce it. Registry and Master Data db-seed images clone this repo at build time, so what is committed here is what a seeded install contains.

Layout

Directory	What it holds
packs/	Country packs — administrative hierarchy, P-codes and boundary geometry, one directory per country. The only supported way geography reaches the platform. Start with its README.
scripts/packs/	Pack tooling: build a real pack from OCHA COD-AB, redraw the fictitious one, and validate either.
demography/	Shared synthetic people: individuals.csv, households.csv, ~500 portrait images. Registry-agnostic — person i0001 is the same individual in every registry, so cross-registry scenarios line up.
household-info/	The same population as consolidated wide CSVs (housing, assets, livelihoods, scores).
users/	Sample portal users.
geo/	Legacy. The pre-pack flat geo.csv. Superseded — see its README.
scripts/	Generators for everything above.

Which country does an install get?

Kamuntu — a fictitious country (packs/XKM) — by default.

That default is deliberate. A demo install attaches invented poverty, enrolment and payment figures to whatever geography it holds, and doing that to a real country's name and outline is not something to do by accident. Real country packs are first-class and one Helm value away (geoSeed.countryPack in the openg2p-master-data chart), and they carry the licence obligations their manifest.json records.

Generators

Nothing here is generated at install time — the outputs are committed, so seed data is reviewable in a diff and reproducible. Regenerate deliberately:

cd scripts/packs
python3 generate_synthetic_pack.py --preset kamuntu --out ../../packs/XKM --version 2026-07-30
python3 validate_pack.py ../../packs/ETH ../../packs/XKM     # always

The pack tools are pure Python — no shapely, no GEOS. Everything else needs pip install -r scripts/requirements.txt.

What does not belong here

Data and generators here must be usable by more than one registry or service. A generator that writes one registry's own tables belongs with that registry, beside the schema it targets.

NSR's bulk sample generator, its distributions.json and the query that rebuilds them used to live here and now sit in national-social-registry/docker/db-seed/. They write NSR's extension tables and draw every enum-backed value from NSR's enums, and keeping them a repo apart from those enums had a concrete cost: a commit reworking the programme enums could not show the generator going stale, so it and reporting_views.sql drifted into agreeing with each other on eleven values the schema never defined. NSR's test/test_enum_values.py guards that now — a test only writable once the two live together.

generate_nsr_data.py and generate_farmer_data.py do stay: they read the shared demography above and write committed fixture files, so they move with demography/'s columns rather than with a registry's schema.