data/wiadomosci_statystyczne_pl/wiadomosci_statystyczne_pl.md

Wiadomosci Statystyczne. The Polish Statistician (GUS) - CC BY-SA 4.0 Polish articles

This slice is derived from the deterministic enumeration of the official journal portal (https://ws.stat.gov.pl/): all articles published from 2022 onward whose landing page carries an explicit per-article "udostepniony na licencji CC BY-SA 4.0" statement and a Polish-language PDF marker.

• Enumerated articles (2022+): 460
• With explicit per-article CC BY-SA 4.0 line: 315
• Source slice retained after QA: 182
• Exact title/DOI overlaps with biblioteka_nauki excluded: 13
• Proposed records: 169
• Proposed tokens: 1,506,269 (cl100k_base proxy)
• Characters: 4,142,041; author coverage: 96.4%
• License: CC BY-SA 4.0 per record
• Immutable source snapshot: https://huggingface.co/datasets/PiotrSty/wiadomosci-statystyczne-pl-articles/tree/078ecdc38e8227faff1769f3ad7f080f44a44bef

Per-record attribution, landing URLs, DOIs, volume/issue, PDF checksums and the verbatim per-article license line are preserved; preserved landing-page snapshots let a reviewer re-check every license statement. The DynaWord proposal excludes all exact normalized-title or DOI matches found against 42,071 pinned biblioteka_nauki rows. Cross-source text deduplication and benchmark checks remain integration gates. pypdf-derived text can retain layout artifacts, math-alphanumeric glyph noise, omitted figures and flattened tables. Per-article licensing does not establish the status of every quoted third-party passage.

What the texts are

• Genre: peer-reviewed statistics-journal content - research articles ("Statystyka w praktyce"), statistics-history articles ("Z dziejow statystyki"), conference reports and reviews ("Dyskusje. Recenzje. Informacje"), obituaries ("In memoriam"). One record per article.
• Register: formal academic written Polish with abstracts (PL/EN), JEL classes, tables and reference lists; bilingual section headings. Author bylines and affiliations are published credits. Not OCR'd (born-digital publisher PDFs), not machine-translated.
• Time span: publication dates 2022-2026 (per-record created; the per-article CC BY-SA 4.0 grant starts 2022-01-01).

Scrubber and residual-PII spot check

Scrubber that ran: pattern redaction in src/build_wiadomosci_statystyczne_pl.py

• e-mail -> [REDACTED:EMAIL], labelled phone (telefon/tel./phone) -> [REDACTED:PHONE]. This is not SlayerLab/NERGAL; the shard must be re-run through NERGAL before it is admitted to a release. Merge with release: None is fine meanwhile.

Residual-PII spot check (deterministic sample of 60 of 169 records, seed 20260928): 96 [REDACTED:*] placeholders remain (the applied redaction); the only residual hits are DOIs in bibliographies and ORCID iDs - public academic identifiers, not PII; 0 hits on all other classes (26-digit runs, IBAN-like, unlabelled IPs, e-mail-like incl. obfuscated forms, street cues, PESEL mentions). Manual reading of 8 records found published credits only.

Normalization (before/after)

src/build_wiadomosci_statystyczne_pl.py: pypdf text extraction -> strip_running_heads() (drops the "Wiadomosci Statystyczne. The Polish Statistician, year, vol., issue, pages" running head and repeated digit-masked page furniture) -> normalize() (NFKC; soft-hyphen/zero-width/control strip; per-line space collapse and trim; pure-number lines dropped; hyphenation across line breaks joined; wrapped lowercase lines joined; blank runs capped) -> PII pattern redaction. Before/after excerpts from a real publisher PDF are in the review thread. Known residue: hyphenation broken by spaces ("Kato - wicach") and pypdf glyph-spacing artifacts survive in some records - already listed under Limitations.