fmbeilin/gus-bdl

GUS BDL — Polish Local Data Bank, full local extract

Complete extract of the GUS Bank Danych Lokalnych (Local Data Bank of Statistics Poland) via its public API: ~1.26 billion observations covering every published variable at every administrative level from gmina (municipality) upward, years 1995–2024 where available.

Files

File	Unit level	Units	Variables	Rows
facts_gminy/part-*.parquet (16 files)	6 — gminy (municipalities)	4,180	35,631	1,051,816,661
facts_powiaty/part-*.parquet (4 files)	5 — powiaty (counties)	382	63,748	154,585,967
facts_podregiony.parquet	4 — podregiony (subregions)	73	67,962	32,782,913
facts_wojewodztwa.parquet	2 — województwa (voivodeships)	16	171,845	17,094,800
facts_makroregiony.parquet	1 — makroregiony (macroregions)	7	171,845	7,478,975
codebook.parquet	—		171,912	variable descriptions
units.parquet	—	4,658		unit directory

No level 0 (national), level 3 (regiony), or level 7 (localities) data is included.

Gminy and powiaty are split into part files by variable_id range so remote readers fetch only the small file a query needs; manifest.json maps each part to its variable range. The other levels are single files.

Schema

Fact tables: variable_id (int), unitId (12-char zero-padded GUS unit code, string), unitName, unitLevel, year, value (double), attr_id (GUS attribute flag).

codebook.parquet maps variable_id to subject codes, human-readable Polish descriptions (variable_full_name), dimension breakdowns (n1–n5, e.g. sex, age group), and measure units. units.parquet lists each territorial unit with its level, voivodeship TERYT prefix, and first/last observed year.

Gminy counts exceed the ~2,477 actual municipalities because urban-rural gminy appear as up to three units (combined, town, rural part).

Querying without downloading

DuckDB reads these files in place over HTTP with predicate pushdown:

SELECT f.year, f.value, c.variable_full_name
FROM read_parquet('hf://datasets/fmbeilin/gus-bdl/facts_gminy/part-*.parquet') f
JOIN read_parquet('hf://datasets/fmbeilin/gus-bdl/codebook.parquet') c
  USING (variable_id)
WHERE f.unitId = '011212001011'   -- gmina Bochnia
  AND c.subjectId = 'P1312';

Python users can also load slices via pandas.read_parquet("hf://datasets/fmbeilin/gus-bdl/facts_wojewodztwa.parquet") or the datasets library using the configs above.

Provenance

Extracted from the official BDL API (bdl.stat.gov.pl/api/v1) December 2025 – February 2026. Values are as published by GUS; no cleaning, imputation, or revision beyond type normalisation (unit IDs zero-padded to the canonical 12-character form). Extraction and packaging pipeline: github.com/fmbeilin/bdl-gus.

Attribution: data source — Główny Urząd Statystyczny (Statistics Poland), Bank Danych Lokalnych, bdl.stat.gov.pl.