Datasets:

---

SlayerLab

/

polish-dynaword

Like 27

Follow

Fabryka AI 97

Datasheet for dziennik_ustaw: what the shard contains (measured), Monitor Polski claim withdrawn, route and Art. 4 wording (v2#107). No data, label or Parquet change.

What changes:

• data/dziennik_ustaw/dziennik_ustaw.md is now hand-written (custom_datasheet: True in the registry, so make_docs.py no longer overwrites it). New "What the shard contains": a table of the 35,442 documents by heading (regulations 18,900; announcements 8,070, of which 7,691 announce a consolidated text; statutes 6,042; Constitutional Tribunal rulings 1,410; government statements 532; other headings 404, mostly agreements, protocols and conventions; resolutions, orders and decisions 45, three of them EU decisions; communiqués 39), the foreign-language treaty texts (307 documents with a Dodatek - Tekst ... w języku X heading, 274 of them English; a stop-word estimate flags 91 documents with 3+ non-Polish blocks, about 0.07% of characters, stated as an estimate), and the date spread (first heading date 1925-2025).
• Monitor Polski is not included. The registry name and the old card said "Dziennik Ustaw + Monitor Polski". The text shows no Monitor Polski item (0 documents cite an own M.P. address in the heading block, 0 Zarządzenie headings). This is an inference from the text; the build kept no publisher field and fetch_eli.py defaults to DU MP, so the flags of the actual run are not recorded. The card says so.
• Legal status (our assessment): normative acts (Art. 4 pkt 1) and other official documents (pkt 2), Polish law only, status may differ elsewhere; public-domain (official documents) is a descriptive term, not a licence; no CC public-domain mark applied or implied; the ELI documentation states no terms and no disclaimer was found. Third-party fragments, the foreign-language treaty texts (status under other states' law) and personal data are outside the statement. The old "reuse unrestricted" sentence and "official drafts" are gone.
• Route and date: ELI interface of the Kancelaria Sejmu directly, src/fetch_eli.py, no intermediary. The download date is not recorded; added 2026-06-15 is the registry's date, not an acquisition date.
• Credit: Dziennik Ustaw; Kancelaria Sejmu RP for the ELI interface. The System Informacyjny Sejmu note is not applied (the SIS terms page does not mention the ELI interface), as in the root NOTICE.md.
• Withdrawn claim: the old card said each act's meta.url is its ELI address. The Parquet has no such field (columns: id, text, source, added, created, token_count, license, author), and the ids dziennik_ustaw_2377725 to _2413166 are a consecutive build counter, not ELI positions.
• src/sources.py (CRLF, dziennik_ustaw entry only): pretty without "+ Monitor Polski", traceable and provenance rewritten to match the card, custom_datasheet: True. Label, SPDX, created, release, added unchanged. README.md (CRLF): the one table row uses the new pretty.
• New src/measure_dziennik_ustaw.py re-derives every figure above from the Parquet. src/test_measure_dziennik_ustaw.py (63 tests, synthetic documents) and src/test_dziennik_ustaw_contract.py (the card matches the registry, README and the shipped shard; the 1-minute language estimate runs only with DU_MEASURE_BLOCKS=1, and was run once: 91 documents, 0.07%).

No sidecar. The build dropped the upstream metadata, so no ELI address, URL or publication date exists in any local file; a per-row sidecar would need a new fetch of about 35,000 acts, which needs approval.

Not changed: per-record license values and the registry label (v2#60), the Parquet, stats, build_dynaword.py. PAPER.md still says "Dziennik Ustaw + Monitor Polski" in four places (lines 20, 101, 142, 179); I did not edit it. The registry created "1918-01-01, 2025-12-31" is a placeholder repeated in every row; the card says so and does not change it (rebuild).

Tests: cd src && python3 -m pytest -q -p no:cacheprovider -rfE gives 1328 passed, 8 skipped, 1 failed. The failure is the known test_studia_bas_contract.py::test_studia_bas_artifacts (parquet missing in this checkout), unrelated. Run on the 4be16d1-based tree; main has since moved only in the uzp_orzeczenia_pl files (PR 152), which this PR does not touch.

Refs slayerlabs/polish-dynaword-v2#107.

🤖 Generated with Claude Code