Datasets:

---

SlayerLab

/

polish-dynaword

Like 24

Follow

Fabryka AI 87

Add OPEN ICM Polish publications pilot

Adds open_icm_pl: 85 records and 2,358,174 measured cl100k_base proxy tokens from a deterministic 100-record pilot of the official OPEN ICM repository.

Source dataset: https://huggingface.co/datasets/PiotrSty/open-icm-pl-publications/tree/c4a20022e6947f55f82c3354bb5c86609ba681b8

What the texts are

Genre: Polish open-access academic publications — journal articles (economics, psychology, pedagogy, law), monographs and edited volumes, conference texts; one record per publication. Register: formal academic written Polish with abstracts, bibliographies and byline blocks (authors, affiliations, ORCID, contact e-mails). Text is repository-generated from PDFs (extraction-noise-prone): math/font glyph noise, hex-encoded UTF-16 fragments and flattened tables survive in places. Time span: issued 1998 – 2024-09 (pilot; 6,183 items eligible in total — yields must not be extrapolated).

License at the origin

• Origin portal: OPEN ICM (Interdisciplinary Centre for Modelling, University of Warsaw) — https://open.icm.edu.pl/home.
• The license statement lives on the portal, not on the mirror: every item carries dc.rights + dc.rights.uri in its origin metadata. Example — https://open.icm.edu.pl/items/0078ca93-45a5-4bef-912d-297c069ee6ab declares dc.rights: "Uznanie autorstwa 3.0 Polska" / dc.rights.uri: http://creativecommons.org/licenses/by/3.0/pl/, readable without JavaScript at https://open.icm.edu.pl/server/api/core/items/0078ca93-45a5-4bef-912d-297c069ee6ab and rendered on the landing page. 77 of 85 records additionally ship a per-item license.txt bitstream, pinned by checksum in attribution.jsonl.
• Allowlisted labels only (fail-closed gate): CC-BY-3.0-PL (32), CC-BY-SA-3.0-PL (38), CC-BY-SA-3.0 (1), CC-BY-4.0 (7), CC-BY-SA-4.0 (7). Dozwolony użytek and NC variants are rejected.

Sample (verbatim, from the built shard)

Journal article:

id: open_icm_pl_232f5743-6863-48e3-a87a-12a2b6a5137c
source: open_icm_pl
added: 2026-09-10
created: 2023
token_count: 11145
license: CC-BY-4.0
author: Wrona, Bartłomiej

STUDIA EKONOMICZNE. Gospodarka • Społeczeństwo • Środowisko

(ECONOMIC STUDIES. Economy • Society • Environment) Nr 1/2023 (11)

− 47 −

Bartłomiej WRONA
Akademia Nauk Stosowanych w Nowym Sączu, Polska

ORCID: 0000-0003-3418-7080

EFEKTYWNOŚĆ I SPOSOBY JEJ POMIARU

ORAZ ROLA PAŃSTWA W JEJ KREOWANIU

Streszczenie

Gospodarka rynkowa wymusza na podmiotach je tworzących takie prowadzenie działalności,

które można uznać za najbardziej efektywne. Jednak samo pojęcie efektywności uznawane jest

za trudne w definiowaniu. Również sposoby pomiaru efektywności nie są jednolicie przedstawiane

w literaturze przedmiotu. Celem artykułu jest przedstawienie efektywności, sposobów jej

mierzenia i roli państwa w podnoszeniu efektywności rynku. […]

Conference text (shows the PDF-extraction glyph noise honestly — the second line's e-mail is mangled by broken font encoding):

id: open_icm_pl_1f4656fd-bcb6-4577-b8b4-82aff31e3873
source: open_icm_pl
added: 2026-09-10
created: 2009
token_count: 14795
license: CC-BY-SA-3.0-PL
author: Lelonek, Bernadeta

PFK: Teksty Pokonferencyjne, 4 (2009), 232–240
Procesy poznawcze a uzalez ̇nienie od grania hazardowego – błe ̨dy mys ́lenia nałogowych hazardzistów dr Bernadeta Lelonek
Instytut Psychologii
Katolicki Uniwersytet Lubelski im. Jana Pawła II
❬❜❡♥✐❛✳❛❧(❡)❡❣♦❅❣♠❛✐❧✳❝♦♠❪
Abstrakt. Problem uzalez ̇nienie od hazardu pomimo tego, z ̇e aktualnie dotyka coraz wie ̨kszej cze ̨s ́ci społeczen ́stwa, nie doczekał sie ̨ jeszcze na gruncie polskim szerzej zakrojonych badan ́ zmierzaja ̨cych do bliz ̇szego poznania jego specyfiki. Nieznajomos ́c ́ problemu skutkuje niemoz ̇nos ́cia ̨ opracowywania adekwatnych strategii prewencyjnych i terapeutycznych. Nałogowe granie wpisuje sie ̨ w kra ̨g uzalez ̇nien ́ czynnos ́ciowych. Podobnie jak inne uzalez ̇nienia powstaje na bazie okres ́lonych mechanizmów. Tym […]

Monograph:

id: open_icm_pl_727d10bb-6b67-4995-9abe-152679f6dadc
source: open_icm_pl
added: 2026-09-10
created: 2016
token_count: 137952
license: CC-BY-3.0-PL
author: Pomianowski, Piotr; Lewiński, Tomasz; Kruszyński, Wojciech

Orzecznictwo organów dyscyplinarnych w wybranych zawodach zaufania publicznego
Piotr Pomianowski
Tomasz Lewiński
Wojciech Kruszyński
MONITORING SĄDOWNICTWA DYSCYPLINARNEGO campidoglio
WYDAWNICTWO

Orzecznictwo organów dyscyplinarnych w wybranych zawodach zaufania publicznego
Piotr Pomianowski
Tomasz Lewiński
Wojciech Kruszyński
Warszawa 2016 campidoglio
WYDAWNICTWO
Tytuł: Orzecznictwo organów dyscyplinarnych w wybranych zawodach zaufania publicznego
Autorzy:
Piotr Pomianowski – rozdziały: I, II, III, IV, VII, VIII, X, XI, XII i XIII
Tomasz Lewiński i Wojciech Kruszyński – rozdziały: V, VI i IX
Recenzenci:
Przemysław Konieczniak
Paweł Skuczyński
Sławomir Stasiorowski
Paweł Dzienis
Korekta:
Katarzyna Rojek
Projekt typograficzny, skład i łamanie:
Marcin Trepczyński
Wydawca:
Campidoglio […]

PII - which scrubber ran

Pattern redaction in src/fetch_open_icm_pl.py (e-mail -> [REDACTED:EMAIL], labelled phone -> [REDACTED:PHONE], ID/IBAN patterns); pipeline counts over the shard: 112 e-mail and 4 labelled-phone redactions. This is not SlayerLab/NERGAL.

Residual-PII spot check (60 of 85 records, seed 20260928) — mixed result, reported precisely:

• 85 [REDACTED:EMAIL] placeholders confirm the redaction works on clean contacts.
• One real residual contact: a glyph-garbled work e-mail in open_icm_pl_6ec903ca-db89-4a17-acc8-850420c468ac (JKolbusz @wsiz.rzeszow.pl, stray space before @ in a line mangled by broken font encoding) that the pattern missed — extraction noise breaks pattern matching.
• Benign, not PII: 161 digit runs (DOIs, URL slugs), 5 26-digit runs (hex-encoded UTF-16 extraction garbage), 1 IPv4-like hit (a CiteSeerX DOI URL fragment), 28 ORCIDs, 25 institutional byline addresses.

Residual risk is moderate for this shard (higher than clean-text sources) — which is exactly why the model-based NERGAL pass before release matters here. Merge with release: None is fine meanwhile. Stated in the datasheet.

Normalization and reproducibility (before/after)

src/fetch_open_icm_pl.py pulls the repository-generated TEXT bitstreams (MD5-verified) and applies Unicode/whitespace normalization, PII pattern redaction, language/quality filtering and exact/near dedup. Raw->shipped before/after excerpts cannot be regenerated anonymously at review time: the OPEN ICM bitstream endpoints serve an interactive CAPTCHA page to anonymous clients — this is what a re-fetch of .../bitstreams/3fd76dd5-.../content returns:

<!doctype html>
<html>
<head>
    <meta charset="UTF-8">
    <title></title>
    <link rel="stylesheet" href="captcha.css">
…
    <div class="captcha-container">
        <div class="captcha-box">

The acquisition documented in attribution.jsonl went through that gate; raw bitstreams are pinned per record by MD5 checksum. The extraction-artifact residue is visible in the samples above.

Rights and method (original)

Each source record carries an allowlisted item-level CC BY or CC BY-SA label. Authors, original item URLs, normalized license identifiers, selected TEXT bitstream metadata and checksums are preserved. The text was fetched directly from official OPEN ICM repository bitstreams after its interactive CAPTCHA, then normalized, checked for language/quality and limited PII patterns, and exact/near-deduplicated within source.

Naming (review item 4)

Renamed per AGENTS.md: src/fetch_open_icm_pl.py (was build_open_icm_pl.py) and src/test_open_icm_pl_contract.py (was test_build_open_icm_pl.py); the test's module import is updated.

Overlap and Slayer ontology

The source pilot retained 95 records after QA. This PR excludes all 10 exact normalized-title or DOI matches found against 42,071 biblioteka_nauki rows at pinned DynaWord revision fe44303ea2db290c7a195264226477f724973357, leaving 85 proposed records. No fuzzy title match reached 0.90. This is metadata evidence, not corpus-wide text deduplication.

The manifest separates content-addressed source and target Versions, Protocol, actual Run, Evidence, falsifiable Claims, Actors and typed lineage. Cross-source text deduplication, benchmark checks, review of quoted third-party passages and controlled training ablations remain pending. This PR proposes a source, not a merged or stable DynaWord release.

Limitations

Repository-generated TEXT can retain PDF extraction errors, omit figures and flatten tables. Item-level licensing does not establish the status of every quoted third-party passage. Pilot yields must not be extrapolated to all 6,183 eligible items.

Test plan

• python3 -m pytest src/test_open_icm_pl_contract.py — 9 passed (after the rename, rebased onto main @ 34e7ea3)
• Residual-PII spot check (above, incl. one real residual contact)
• License verified at the origin (portal item metadata + license.txt bitstreams)
• Reviewer: mixed per-record CC BY / CC BY-SA treatment accepted
• Reviewer: NERGAL-before-release note accepted

@kacperwikiel Ready for review. This PR is based on current main (fe44303) and shows no merge conflicts.

Data samples are provided the same way as other sources:

• In this PR: data/open_icm_pl/open_icm_pl.sample.jsonl — 12 deterministic records with all metadata columns (id, text, source, added, created, token_count, license, author).
• The open_icm_pl.parquet ships in-tree (LFS), so the dataset preview renders.
• In the pinned immutable source dataset: artifacts/sample.jsonl (linked in the PR description) at PiotrSty/open-icm-pl-publications@c4a20022.

85 records / 2,358,174 cl100k_base proxy tokens after excluding the 10 exact title/DOI overlaps with biblioteka_nauki. Thanks!

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

1. Sample inline. Paste 2–3 real documents into the description, one fenced block each: metadata as key: value lines, a blank line, then the text with its real line breaks, truncated with […] (format as in #24). A linked .sample.jsonl renders on the Hub as one escaped line per document and can't be reviewed at a glance.
2. License at the origin. Link the license statement on the OPEN ICM portal, not only your HF mirror — the mirror can't be evidence for itself. Link the origin portal too.
3. Describe the texts — genre, time span, register.
4. Naming. Use src/fetch_<key>.py and src/test_<key>_contract.py instead of build_* / test_build_*.
5. PII. AGENTS.md requires SlayerLab/NERGAL. The source can merge with release: None after another scrubber, but it won't be admitted to a release until the shard is re-run through NERGAL. State in the datasheet which scrubber ran.

Review addressed — rebased onto main @ 34e7ea3 (the src/sources.py conflict is resolved; the entry stays with release: None). Point by point:

1. Sample inline. 3 real documents in the description, one fenced block each (key: value metadata, blank line, real line breaks, […] truncation, format as in #24): a journal article, a conference text (shown with its real PDF-extraction glyph noise — honest), and a monograph. The metadata block carries the per-record license so the mixed CC BY / CC BY-SA treatment is visible.
2. License at the origin. Now cited at OPEN ICM itself: origin portal https://open.icm.edu.pl/home, an example item landing page (https://open.icm.edu.pl/items/0078ca93-45a5-4bef-912d-297c069ee6ab) with its dc.rights: "Uznanie autorstwa 3.0 Polska" / dc.rights.uri: http://creativecommons.org/licenses/by/3.0/pl/ statement, machine-readable without JavaScript at https://open.icm.edu.pl/server/api/core/items/0078ca93-45a5-4bef-912d-297c069ee6ab, plus the per-item license.txt bitstreams (77 of 85 records, checksum-pinned in attribution.jsonl). The label mix is spelled out (CC-BY-3.0-PL ×32, CC-BY-SA-3.0-PL ×38, CC-BY-SA-3.0 ×1, CC-BY-4.0 ×7, CC-BY-SA-4.0 ×7) with the fail-closed allowlist. Datasheet updated.
3. Texts described. Genre (journal articles, monographs/edited volumes, conference texts), register (formal academic Polish with byline blocks — ORCIDs, contacts — and PDF-extraction noise), time span (issued 1998 – 2024-09; pilot of 100 of 6,183 eligible items).
4. Naming. Done: src/fetch_open_icm_pl.py and src/test_open_icm_pl_contract.py (module import updated); python3 -m pytest src/test_open_icm_pl_contract.py — 9 passed.
5. PII. Datasheet names the scrubber (pattern pass in src/fetch_open_icm_pl.py, 112 e-mail + 4 labelled-phone redactions — not NERGAL) and reports the residual spot check honestly: one real residual contact survives — a glyph-garbled work e-mail in open_icm_pl_6ec903ca-… (JKolbusz @wsiz.rzeszow.pl) that the pattern missed because extraction noise broke the line (plus benign DOIs/ORCIDs/hex garbage). Residual risk is moderate for this shard — exactly why no release before a NERGAL (model-based) pass; merge with release: None is fine meanwhile.

One negative finding to flag: raw → shipped before/after excerpts could not be regenerated at review time — the OPEN ICM bitstream endpoints serve an interactive CAPTCHA page to anonymous clients (the interstitial is pasted in the description). The raw bitstreams are MD5-pinned per record in attribution.jsonl, and the normalization is fully documented in the committed src/fetch_open_icm_pl.py + datasheet.

Verification: python3 -m pytest src/test_open_icm_pl_contract.py — 9 passed on the rebased branch.