Datasets:

SlayerLab
/

polish-dynaword

Fills author and adds data/wolne_lektury/wolne_lektury.attribution.jsonl, one line per row (6,076), in row order, joined by id. The text is unchanged: the builder checks each row's SHA-1 and a whole-table digest. Only the author column of the Parquet changes.

author: 6,039 rows, from Wolne Lektury's own book headers. Translations read "Autorzy oryginału: A; Tłumaczenie: T" (the global_voices form). 37 are empty: WL's placeholders "Autor nieznany" (36) and "Praca zbiorowa" (1) with no named translator. Names are WL's, turned to "Given Family"; pseudonyms and cryptonyms are kept as WL gives them.
Sidecar: title, authors, translators, work link, the full licence URL of the exact version, every person and note WL credits, the print edition (6,032 rows), the line "Książka pochodzi z serwisu Wolne Lektury" and a processing note. It covers every row, not only the 1,517 licensed ones; it restores the credits of the export footer that is not in the text.

Licence	Rows	`license_url`
public domain	4,559	none
LAL 1.3	1,467	`http://artlibre.org/licence/lal/pl/`
CC BY-SA 3.0 PL	34	`https://creativecommons.org/licenses/by-sa/3.0/pl/`
CC BY-SA 3.0	5	`https://creativecommons.org/licenses/by-sa/3.0/`
CC BY-SA 4.0	11	`https://creativecommons.org/licenses/by-sa/4.0/`

For the 34 "3.0 PL" rows WL's rights text says 3.0 PL but its link is the 3.0 Unported URL. license_url follows the text; WL's link is kept in wl_license_url.

Checks (in `wolne_lektury.license-evidence.json`)

Normalisation: in this repository the only change to a text is trimming leading and trailing whitespace; no paragraph is removed. What SpeakLeash did before is not documented, and who removed the export footer is not established. The card says so.
Notes and prefaces: all 6,076 texts scanned, 0 with WL credit text; 40 random rows read at both ends. Headings found: Wstęp 73, Przedmowa 50, Od tłumacza 43, Komentarz 10, Posłowie 10 and a few others.
Edition flags, nothing excluded: of the 4,559 public-domain rows, 2,421 carry a flag, e.g. 58 name a critical or scholarly edition and 1,311 follow a print from after 1955. Across all 4,559 public-domain rows, WL's own public-domain date (which counts translators) is 2026 or earlier for 4,529 and missing for 30.
Not run: the footnote check, which needs WL's book XML. Not established: that names follow the first edition, and which edition each modernised text follows.

Requests: only wolnelektury.pl, 6,117 sequential requests at least 1.2 s apart (API book list, book records, and Range requests for the XML headers), about 41 MB; robots.txt allows these paths. No bot checks.

Files

data/wolne_lektury/wolne_lektury.{parquet,attribution.jsonl,license-evidence.json,md,stats.json}
src/attribution_common.py (shared with the eltec_pol builder), src/build_wolne_lektury_attribution.py, src/clean_wolne_lektury.py (docstring), src/sources.py
Tests: src/test_attribution_common.py, src/test_wolne_lektury_attribution.py, src/test_attribution_release.py
artifacts/source_findings.md: WL section