# dziennik_ustaw Dziennik Ustaw (Polish primary legislation) ## Dataset description - **Source (upstream):** https://api.sejm.gov.pl/eli/acts - **Domain:** legal - **Language:** Polish (pl) - **License:** `public-domain (official documents)` - **Created (range):** 1925-04-29, 2024-12-30 - **Added:** 2026-06-15 ## What the shard contains The Dziennik Ustaw (Journal of Laws) only. **Monitor Polski is not included.** The build kept the text and nothing else (the Parquet columns are `id`, `text`, `source`, `added`, `created`, `token_count`, `license`, `author`), so there is no type, publisher or address column; the ELI address of each row is in a sidecar file (see Provenance). The figures below are read from the text: the first line of a document names its type. They are reproduced by `python3 src/measure_dziennik_ustaw.py` on the shipped Parquet. | Heading (first line) | Documents | |---|---:| | Rozporządzenie (regulation) | 18,900 | | Obwieszczenie (announcement); 7,691 of them announce a consolidated text | 8,070 | | Ustawa (statute) | 6,042 | | Wyrok (ruling); all 1,410 are rulings of the Constitutional Tribunal | 1,410 | | Oświadczenie (all 532 are `Oświadczenie rządowe`, government statements) | 532 | | Other headings (treaties and similar instruments, see below) | 404 | | Uchwała, postanowienie, decyzja; 3 are EU decisions (2 of the Council of the EU, 1 of the European Council) | 45 | | Komunikat (communiqué) | 39 | | **Total** | **35,442** | - The 404 other headings start, most often, with `umowa` (132), `porozumienie` (55), `protokół` (54), `dokument` (29), `konwencja` (25), `zmiany` (19), `poprawki` (19), `program` (14), `poprawka` (7), `zmieniony` (7), `traktat` (6) and `układ` (5): international agreements and related instruments (protocols, conventions, amendments, executive programmes). Misspelt headings in the data (`Rozporzadzenie`, `Roporządzenie`) are counted with their type. - **Monitor Polski.** The registry and an earlier version of this card said "Dziennik Ustaw + Monitor Polski". Every row is linked to its own Dziennik Ustaw act (see the sidecar under Provenance), so no row is a Monitor Polski item. The counts agree: the Dziennik Ustaw lists for 1990-2025 hold 35,444 acts with an HTML text (none of them from 2025), which is the 35,442 rows plus two acts that have no row. The text agrees too: no document cites itself with a Monitor Polski address (`M.P.` year, `poz.`) in its heading block, and no heading begins with `Zarządzenie` (0 of 35,442). - **Repealed acts are included.** The usage line of `src/fetch_eli.py` passes `--in-force-only`, but the ELI lists carry no `inForce` field, so the flag keeps every act; all listed acts with an HTML text are in the shard, whether in force or not. - **Dates.** `created` is the act's own date, per row (1925-04-29 to 2024-12-30; before v0.2.6 every row held the registry's range `1918-01-01, 2025-12-31`). It comes from three sources, each with errors: the first "dnia" date in the heading block, the first "z dnia" date in the ELI title, and the list's `announcementDate` (despite its name, the act's date as keyed). A date after the act's publication is dropped as impossible, and the date at least two sources agree on is taken (35,299 rows). Otherwise the heading's date is taken (141 rows, nearly all international agreements, whose heading gives the date of signature), else the list's (1 row); one act, `DU/2024/1647`, whose three sources all give 5 November 2025, gets its publication date (2024-11-12). The sidecar records which sources gave each date. Measured from the text alone, the first date in the heading block lies between 1925 and 2025 (36 documents before 1990; 1,347 in the 1990s; 3,360 in the 2000s; 18,316 in the 2010s; 12,383 in the 2020s). - **Foreign-language text.** Treaty texts appear in the other authentic languages: 307 documents carry a heading of the form `Dodatek - Tekst ... w języku ` (misspelt headings are not counted). The language is English in 274 documents (271 spelled `angielskim` and three variant spellings) and French in 18; 43 further spellings name other languages, among them Ukrainian (8 documents), German (7), Russian (7), Arabic (6) and Czech (6). A stop-word estimate over blocks of 15 lines finds at least three non-Polish blocks in 91 documents, about 0.07% of all characters. That figure is a heuristic estimate, not the output of a language detector. ## Licensing — traceable basis **Legal status (our assessment).** We consider the texts in this shard normative acts (Art. 4 pkt 1 of the Polish Copyright Act) or other official documents (Art. 4 pkt 2: official documents, materials, signs and symbols), and therefore excluded from copyright under Art. 4 (Act of 4 February 1994, consolidated text Dz. U. 2025 poz. 24; text: https://eli.gov.pl/api/acts/DU/2025/24/text/O/D20250024.pdf). That is a statement of Polish law only; their status may differ in other jurisdictions. The label `public-domain (official documents)` is a descriptive term for this status, not a licence: no Creative Commons public-domain mark (CC0, Public Domain Mark) is applied or implied, and no upstream Creative Commons license is asserted. The ELI documentation (`/eli.html`) states no licence or terms of use, and we found no disclaimer on it. The statement does not extend to third-party fragments inside the documents, nor to the foreign-language treaty texts (part of the official publication, but their status under the law of other states is outside this statement), and personal data is a separate matter: this card records no personal-data pass for this source. The Polish Open Data Act (Dz. U. 2023 poz. 1524, articles 2(12), 5, 6, 14, 15 and 17; text: https://eli.gov.pl/api/acts/DU/2023/1524/text.html) is cited as the reuse framework for the institutions' own material, not as a basis for third-party works. This is our assessment, not legal advice. **Credit:** Dziennik Ustaw; Kancelaria Sejmu RP for the ELI interface. We do not apply the System Informacyjny Sejmu source note to this source: the terms page does not mention the ELI interface. The root `NOTICE.md` has the details. ## Provenance and acquisition - **Route:** the ELI interface of the Kancelaria Sejmu directly, no intermediary. The script `src/fetch_eli.py` lists the acts of a publisher and year, downloads `text.html` of each act that has an HTML text (acts without one are skipped) and reduces it to plain text; the build drops texts under 200 characters (none dropped). It is a native source, not redistributed via SpeakLeash. - **Acquired:** the download date is not recorded and the fetch run is not logged here. The shard's `added` date, 2026-06-15, is the registry's date, not a measured one. The ELI lists used for the sidecar below were read on 2026-10-19. - **Per-row ELI address (sidecar).** An earlier version of this card said that each act's `meta.url` is its ELI address. The build dropped that field, and the ids (`dziennik_ustaw_2377725` to `dziennik_ustaw_2413166`, consecutive) are a running counter of the build, not ELI positions. `dziennik_ustaw.attribution.jsonl` restores the address: one line per Parquet row, in the same order, with `id`, `eli` (e.g. `DU/1990/541`), `source_url` (the act's ELI API address), `match`, `published` (the list's promulgation date; null for 404 rows), `created`, `created_from` and `text_sha1`. `src/build_dziennik_ustaw_attribution.py` builds it from the shard and the ELI year lists of Dziennik Ustaw for 1990-2025 (pinned by SHA-256), without fetching any act text. `src/fetch_eli.py` wrote the texts in list order and the build cut them into chunks of 2,000 lines, so a row is linked either by its title (34,089 rows: the heading equals or starts with the ELI title, on the same date, with one candidate) or by its position between two title-linked rows (1,353 rows, among them acts whose list title has changed since publication, headings with typos, Constitutional Tribunal rulings of the same day and same-day acts with the same title). A gap whose row count differs from the number of acts between its two ends stops the build; none did. As a spot check, of the 1,353 position-linked rows 1,303 agree with their act on the heading's date and first word, 35 on the first word only, 13 on the date only and 2 on neither (misspelt headings, and treaty headings that put the instrument's type first). The list titles are the acts' current titles, so the sidecar does not repeat them. - **Two listed acts have no row:** `DU/2005/1538` and `DU/2008/59`. `src/fetch_eli.py` drops an act whose text request fails four times or whose text is under 200 characters; which of the two happened is not recorded. - A new fetch returns what the API serves on the day, so it does not reproduce this shard byte for byte. ## Statistics | documents | characters | tokens (tiktoken proxy) | |---:|---:|---:| | 35,442 | 1,226,137,838 | 486,126,575 | ## Per-document license metadata | license | documents | |---|---:| | `public-domain (official documents)` | 35,442 | Author metadata present for **0** documents. Empty values mean the upstream record did not expose a machine-readable author field. Statistics were recomputed directly from the released parquet file. ## Filters applied (build_dynaword.py) Minimal, per Dynaword guidelines (heavy filtering left to downstream use): - drop documents < 200 chars: **0** - drop non-Polish (diacritic ratio): **0** - exact cross-source dedup (sha1): **0** - OCR alpha-ratio < 0.70 (OCR sources only): **0** - read 35,442 → kept 35,442 Token counts are a fast tiktoken (cl100k) proxy (~1% off Llama-3); the canonical Llama-3 count is computed at release.