Datasets:

---

SlayerLab

/

polish-dynaword

Like 24

Follow

Fabryka AI 87

What

Wording fixes for 1000_novels. No data, stats or parquet change.

• data/1000_novels/1000_novels.md: the card said the CLARIN-PL record "licenses the corpus under CC BY 4.0 as a compiled resource". It now says the record states CC BY 4.0, but the handle holds metadata only, so no licence covers the texts.
• data/1000_novels/1000_novels.license-evidence.json: upstream_record_license drops "compilation" and says the CC BY 4.0 is stated on a metadata-only record and does not license the texts.
• README.md and src/make_docs.py: the release-status bullet still read "included since v0.2.1 under the compiled corpus's CC-BY-4.0 licence…". It now describes the 837-text public-domain subset (Hub PR 49) and the metadata-only record.

Why

CLARIN-PL (Tomasz Naskręt, email reply 2026-10-05) said that handle 11321/312 does not provide the corpus files and holds no documentation of rights in the individual texts. A CC BY 4.0 on a record with no files cannot be read as a licence for the compiled corpus. The release keeps only public-domain texts, so it never relied on that licence. Only the description was wrong.

Checks

• pytest src/ in the full checkout with these edits: 281 passed.
• In this LFS-less worktree, results are the same before and after the change. The 30 data-contract tests that need the parquet files fail in both runs.
• No render_sample.py output: the text and rows are unchanged.

🤖 Generated with Claude Code