The IASB notice in these documents allows reproduction within the EEA and
reserves all rights outside it; the Hub is not EEA-limited. Maintainer
decision 2026-10-06 (slayerlabs/polish-dynaword-v2#57).
src/clean_eurlex.py: clean() relabels and drops the ids pinned in
eurlex.third-party-scan.json. It fails closed, leaving the parquet as it
was, if a row carries the notice without being pinned, a pinned row lacks
it, or a pinned id is missing from the input. --scan refuses to change a
pinned id list.
Rebuilt from the v0.2.5 input (sha256 5f151d57...): 5,976,949,249 ->
5,902,889,135 chars, 2,378,055,718 -> 2,350,709,056 tokens. Every kept row
is identical to v0.2.5 apart from the license column; the rebuild is
byte-reproducible (sha256 17e4112c...).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
eurlex: relabel to LicenseRef-EU-Reuse-Policy, report third-party (IASB) material (#71)
- wikimedia: per-page licence labels for wikinews, wikivoyage and wikibooks (shared code) (cf0470c6ba38a2cbf6e03ffc716df50ec139ab70)
- wikinews: label each page CC-BY-2.5 or CC-BY-4.0 by its last revision (64d2affe44fa3166d6f124a00dd5ed781818126d)
ted_pl: cite both licence bases, rescrub missed e-mails and phones (v0.2.6, v2#57) (#75)
- ted_pl: cite both licence bases, rescrub missed e-mails and phones (v0.2.6, v2#57) (342f9f7d560fa202aab674ac33c020a55884ce2a)
- ted_pl: rebuild text from the TED API with the fixed PII patterns, drop the rescrub step (v0.2.6, v2#57) (a28e3cffdd36ddeb48603081b839bd3a1158cdbc)
sejm_api_2011_2022: relabel provenance, drop speeches contained in parliamentary / sejm_api (118,902 -> 32,460) (#76)
- edukacja_medialna_pl: drop 4 lessons with embedded third-party quotations (70c0a6a5b6bb253a6cd09a5cf74c3e3d9a138213)
- edukacja_medialna_pl: datasheet Reproduction section includes the clean step (d0ef0ad6b4e9abdf4227df9490185a8021bc7b9c)
- edukacja_medialna_pl: narrow the clean script's restore command to the six pinned files (7316f59f246d3056befe27db7e6e62e7c08d6d8e)
nonsa: drop namespace 100 Cytaty and <poem> pages (12,304 -> 11,692 docs, v0.2.6 rights audit) (#73)
- nonsa: drop namespace 100 Cytaty and <poem> pages (12,304 -> 11,692 docs, v0.2.6 rights audit) (174ea3429e0cb5faa652f89e75bc6ad9ef271b15)
- nonsa: record why the 99 pages are gone from the live wiki (quality or relevance, none for rights) (98c29e779aa9012aebec6afdf33a14054b778d6f)
open_agh_chemistry_pl: pin the four rights pages in a licence-evidence file (no data change) (#72)
- open_agh_chemistry_pl: pin the four rights pages in a licence-evidence file (4e175abfd35bb577fdde70a142c3b4e1acfee5df)
- Merge origin/main (PRs 67, 70, 68, 69) into open_agh_chemistry_pl evidence (3bfbdd47958a5576ba37e807b32f9f3096e084fa)
sejm_api: state the contributor snapshot in provenance (v0.2.6, v2#57) (#69)
- sejm_api: state the contributor snapshot in provenance (v0.2.6, v2#57) (95e01b92b8929107be2c381e8e47302b0417f9b6)
- sejm_api: drop the stale parlamint_pl sentence from the card and its template (25e6ecc2481c998e41a8d88566628dd1f9baa90b)
- Merge origin/main (PRs 67, 70, 68) into sejm_api provenance (c9a2d80bcbf0e08cbdfef39f00e98d195c669bf4)
wikipedia: rebuild from the official plwiki dump, CC BY-SA 3.0/4.0 per page (v0.2.6) (#68)
- wikipedia: rebuild from the official plwiki dump, label 3.0/4.0 per page (3393e05c178bb29cd9d0bda0444ef9fb2573fbf3)
- wikipedia: drop the two copied pages and copyright-reason {{ek}} pages (ca8d210c9e7f738b89c60bcf3623e31a01d2e868)
- Merge origin/main (wolne_lektury PR 67, 1000_novels PR 70) into wikipedia rebuild (3beca40c58c76404c621fcccd0d2fdb2a074cdfd)
1000_novels: drop texts already in wolne_lektury or eltec_pol (887 → 221 docs) (v2#45) (#70)
- Drop partner republications from global_voices (2,040 -> 1,926) (d6bd578858f0b398899c75379511ce712e226e54)
- Merge main (#64) into global_voices: keep all source_findings entries (438fac3f3b3a17cd10d2204490eac69049cc47bf)
- govpl: drop 188 PAP reprints (gov.pl third-party carve-out) (f9e9eee620a1b15c277f4897d18466228e2c52c7)
- Merge main (#62) into govpl: keep both source_findings entries (a9895188ff7a138d1195865151243854b039b30a)
biblioteka_nauki: drop the 3 Public Domain Mark records (#62)
- Withdraw fandom_pl: Fandom ToU bars scraping and AI training (9893ad9c795bdfd241c3a0061095bcdf01025fa9)
- Merge main into fandom_pl withdrawal (PR #57) (82dbf0530d2f3cbd2130d314b77ec34dfb0d627e)
Withdraw openstax_pl: OpenStax Poland states NC and an LLM-training ban (#56)
- nkjp1m: relabel CC BY 3.0 PL per IPI PAN, add attribution file (400f934ae256bdb9d28da87dbee37435cdf90c46)
- nkjp1m: relabel CC BY 4.0 per nkjp.pl and the consortium's reply (75d65e6d76b1eb44c60aa001a06337a1e183b6e9)
- nkjp1m: record both licensor emails by sha256 (359f3f990bad5ace68c3e208c669329c1fa90036)
- docs: link PPC to its IPI PAN page in the card disclaimer (2785cac14ff778190a0371e2e7ed3f7e69b0cb25)
- nkjp1m: match ATTRIBUTION.md to the text the consortium approved (35a4a7ecff6e05d1b528653184a15630835a7d17)
- nkjp1m: drop the approval quote from the licence evidence (e6bbea723c5e610052aa77af1dd0e6ef338af31d)
wiktionary_examples: drop quoted third-party lines, relabel CC BY-SA 4.0 (#50)
- wiktionary_examples: drop cited example lines (WIP) (049bad406b29734bb64974ece7cb13b3aeaa65f3)
- wiktionary_examples: drop quoted third-party lines, relabel CC BY-SA 4.0 (b1c2a783f3b46acfa35db266feb992d941888adf)
- wiktionary_examples: record the decision to keep the 200-char gate (c158705cab894f72d98427eb3a93b226c3fc0536)
- wiktionary_examples: web sample of unmarked quotations (0/100) (615ef2d849b8ee0d44da8d205eeafa84bb3b50d5)
1000_novels: keep only the public-domain texts (#49)
Brings 7ae4ac7, the new target of the v0.2.5 tag, into main's history.
Conflicts in README.md and src/sources.py resolved by applying the same
change to main's hand-maintained README and CRLF sources.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
docs: move the 1000 Novels rights warning from Excluded to its datasheet
The card listed 1000_novels among the released sources and, at the same time,
in the "Excluded sources" table with a warning that the novels may include
in-copyright works. The source has been released since v0.2.1, so the row
contradicted the source table; the warning itself was correct.
- src/sources.py: drop 1000_novels_corpus_CLARIN-PL from EXCLUDED
- data/1000_novels/1000_novels.md: rights note with the per-text screen
(830 public domain by term, 10 Wolne Lektury public domain, 17 in copyright
under CC BY-SA 3.0, 143 unverified) and the fixes planned for the next release
- src/make_docs.py: README status bullet pointing to that note
- README.md: regenerated with src/make_docs.py (CHANGELOG stub discarded)
Parquet files are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
19 days ago
Docs refresh: merge the re-pointed v0.2.5 tag; document main since v0.2.5
Merges f71cc09 (the commit the v0.2.5 tag now points at) so main agrees with
the published tag: v0.2.5 is 20 sources, 4,366,651 documents / 9,676,513,980
proxy tokens, with sejm_api and wiktionary_examples released.
On top of that:
- README: callout for main (36 sources, 4,094,255 docs / 8.38B tokens);
european_hplt_v3_pl marked withdrawn (dead datasheet link removed);
samorzad_gov_pl v0.2 rebuild noted; table of the 17 sources not yet
released (1,331,877 docs / 1,262.4M tokens), generated from the registry
and stats.json; contributors updated from Hub PR authors; phrase-frequency
section labelled as the v0.2.5 snapshot.
- CHANGELOG: new Unreleased section (17 sources, HPLT removal, samorzad v0.2,
tooling PRs, tag re-point). Release sections untouched.
- src/sources.py: rock_pollub_pl gets its added date; comment on the
withdrawn HPLT entry.
- AGENTS.md: two known-drift notes on regenerating the README from main.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
