Datasets:

---

SlayerLab

/

polish-dynaword

Like 29

Follow

Fabryka AI 101

sejm_interpellations quotes letters from the public, some with health details, and only the pre-NERGAL regex scrub has run on it. It has release: None and is not in v0.2.6, but main's default config (data/*/*.parquet) still serves it through load_dataset and the viewer. Main is as public as a release tag, so the data comes off main until v0.2.7, as uzp_orzeczenia_pl's did in #199.

• Removed: data/sejm_interpellations/ (shard, attribution sidecar, licence evidence, stats and datasheet). It comes back with the shard.
• Kept: the registry entry (release: None), src/fetch_sejm_interpellations.py, src/build_sejm_interpellations_attribution.py and both tests. The tests skip while the folder is missing (src/conftest.py, and the attribution test's own guard).
• src/sources.py: the legal note says the data is held off main, and why.
• README.md: main is now 28 sources, 3,394,883 documents, 7.04B tokens; the "On main, not in v0.2.6" table loses its row and the note under it names the source as held off main.
• CHANGELOG.md: the Unreleased section and a "Removed from main" entry under v0.2.6.
• NOTICE.md: the source's section is removed; the file now covers five sources.
• AGENTS.md (known drift), artifacts/source_findings.md (new entry), docstrings in src/conftest.py and src/make_docs.py.

Older revisions of main still hold the files; removing them from the history is a separate step.

python3 -m pytest src/: 2179 passed, 27 skipped.

🤖 Generated with Claude Code