Datasets:

---

SlayerLab

/

polish-dynaword

Like 24

Follow

Fabryka AI 84

Ontology alignment update is ready for review.

• No dataset rows or binary data artifacts were changed.
• A full rebuild from the pinned source reproduced the existing Parquet and attribution-sidecar SHA-256 digests.
• The manifest now includes content-addressed Object/Version and Protocol/Run records, typed lineage, Actors, addressable Evidence, falsifiable Claims, Claim-Evidence links, and evidence-backed attestations.
• Contract coverage now checks immutable commit URIs, graph references and acyclicity, protocol code identity, environment capture, and artifact checksums.

Verification result: 4 passed. Please let me know if you would like any ontology field names adjusted before merge.

Review against AGENTS.md ("What the PR must show", "Normalize the text before you build"). Needed before merge:

1. License citation. Cite Polish Copyright Act art. 4 pkt 2 with a link in the description (as in sejm_api), not only in .license-evidence.json.
2. Normalization and PII. State which cleaning steps ran and whether NERGAL ran (interjections can name private persons); add a before/after excerpt.
3. Filtering beyond the date-boundary dedup — min length, language ID, or confirm only the standard build gates applied.
4. Shared code. fetch_sejm_api_2011_2022.py looks like it duplicates fetch_sejm_api.py — factor out the shared parsing instead of a second copy, and don't modify make_docs.py.
5. Rebase onto current main. git fetch of this PR's ref currently fails on our side; a fresh push after the rebase should fix it.

Correction to my review comment above: the git fetch failure was on the Hub side (have/ack negotiation), not a problem with this PR. It fetches fine with git -c fetch.negotiationAlgorithm=noop fetch. No re-push needed for that; the rebase ask still stands.

Review addressed — pushed onto refs/pr/18 (this is PiotrSty's PR), rebased onto main @ 34e7ea3. Point by point:

1. License citation in the description. public-domain (official documents) / LicenseRef-Polish-Official-Documents — official parliamentary documents and materials are excluded from copyright by Polish Copyright Act art. 4 pkt 2 (https://eli.gov.pl/api/acts/DU/2025/24/text/O/D20250024.pdf); reuse is documented under the Polish Open Data and Reuse of PSI Act, arts. 2(12), 5, 6, 14(1), 15, 17 (https://eli.gov.pl/api/acts/DU/2023/1524/text.html). Same basis as sejm_api; statutory evidence also rides in sejm_api_2011_2022.license-evidence.json (that file stays — the description now carries the citation itself).
2. Normalization and PII. Which cleaning ran, in order (now in the datasheet): the pinned source release protocol (PiotrSty/sejm-speeches-corpus@1245577, README steps 3–6): terms→term normalization + term/date verification, text-stat recomputation with heuristic PII redaction (e-mail, PESEL, phone) — release 1.0.1's manifest records the actual counts: 1 e-mail replaced, 0 PESEL, 0 phone across the corpus — speeches under 50 words dropped, exact dedup on term + date + speaker + text. Then this build: date window 2011-01-01..2022-12-31, len(text) >= 200 + non-empty author, exact text dedup — no further text rewriting. NERGAL did not run: the scrubber is the upstream heuristic, and floor interjections ("Głos z sali", heckling) can name private persons — so no release before a NERGAL pass (AGENTS.md), merge with release: None fine meanwhile. Before/after: at text level this build is a deliberate no-op (the pinned release keeps post-normalization text + the documented protocol; raw stenogram pages aren't pinned and drift on sejm.gov.pl), so the concrete before/after evidence is the redaction record above (the one e-mail → placeholder) plus the verbatim speech samples already in the description.
3. Filtering — answered precisely. Beyond the date-boundary filter and dedup, this build applies exactly: minimum 200 characters, non-empty author, exact text dedup (sha1). No language-ID gate runs here (the pinned release's QA covers the corpus); nothing else is dropped. Counts in sejm_api_2011_2022.stats.json: 118,902 published of the pinned 2011–2022 slice.
4. Shared code factored out. New src/sejm_api_common.py holds the previously byte-identical copies (sha256_file, canonical_sha256, write_json, EXPECTED_FIELDS semantics, legal_evidence — which differed only in signature — and stable_document_id/stable_id, same body). Both fetch_sejm_api.py and fetch_sejm_api_2011_2022.py now import from it via thin name-preserving wrappers; the second copy is gone. make_docs.py is untouched. The manifest's pinned code_sha256 is updated to the refactored script digest.
5. Rebase — done (was 3 commits behind; drift gone).

Verification: python3 -m pytest src/test_sejm_api_contract.py src/test_sejm_api_2011_2022_contract.py — 16 passed on the rebased tree (after the digest pin refresh).