--- license: other license_name: ted-notice-reuse-terms language: - pl task_categories: - text-generation configs: - config_name: default data_files: - split: train path: data/ted_pl/ted_pl.parquet --- # TED Polish procurement notices Polish narrative text rebuilt from official TED procurement-notice records. - Indexed notices: 20,588 - Search API records: 20,588 - XML fallbacks: 1,255 - Retained documents: 15,544 - Tokens: 42,473,050 (`cl100k_base` proxy) - Dates: 2023-02-15 to 2024-05-15 - Contracting-authority attribution: 100.0% The pinned PleIAs mirror is an identifier index only because its Polish preview contains replacement-character encoding damage. Released text comes from official TED Search API narrative fields or, when absent, official XML; any remaining U+FFFD record is rejected. Structured contact/address fields are omitted; emails, phone numbers and labelled national identifiers in narrative fields are replaced. Recurrent boilerplate and within-source duplicates are filtered. See `NOTICE.md`, `ted_pl.attribution.jsonl`, `ted_pl.decisions.jsonl`, `ted_pl.qa.json`, `ted_pl.sample.jsonl` and `artifacts/ted_pl_ontology_manifest.json` for attribution, selection decisions, QA, full samples and the Slayer ontology manifest. The source index and the checksums file are not in this repository. ## License basis Two bases, recorded separately. Neither is signed off by a lawyer. Clause text, URLs, retrieval dates and page hashes are in `ted_pl.license-evidence.json` (slayerlabs/polish-dynaword-v2#57). **1. EU: TED legal notice** (, read 2026-10-05, sha256 of the page as read `c010943c01af22cf9cb30a32ca4044a3a1ee4e36a090e2dd2d8ee3d15a498997`). Clause relied on: "Unless otherwise noted, the procurement notices published in the Supplement to the Official Journal of the European Union can be freely reused, for commercial or non-commercial purposes." The notice names **Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents** as the Commission's reuse policy (, read 2026-10-05, sha256 `2f1e7b602f21e286dfc576cc9f1387878e1099f9b2c7dcca5777fcbed611e1ed`). The Decision covers Commission documents (arts 1 and 2(1)); it is not shown to license notices drafted by national contracting authorities. The notice's CC BY 4.0 sentence covers SIMAP editorial content and its CC0 sentence covers system metadata; neither is the grant for notice text, so the text is recorded as `LicenseRef-TED-Notice-Reuse` and not relabelled CC0 or CC BY. The notice also says that identifiable private individuals and third-party works may need additional clearance. **2. PL: notices drafted by Polish contracting authorities** (inferred, not verified against a ruling or an authority's statement): - Ustawa o prawie autorskim, Dz.U. 2025 poz. 24 (, read 2026-10-05, sha256 `1c2e9a082a66fad6e26bba1d4d01a22bd67871657416729cea62b4b9f71cff86`), art. 4 pkt 2: "Nie stanowią przedmiotu prawa autorskiego: ... 2) urzędowe dokumenty, materiały, znaki i symbole". - Ustawa o otwartych danych i ponownym wykorzystywaniu informacji sektora publicznego, Dz.U. 2023 poz. 1524 (, read 2026-10-05, sha256 `54820a69c1b4e470cb372a22249558241bc68058ce447abbddfba1c61dd88f05`), art. 2 pkt 8 (public-sector information), art. 3 (obliged entities), art. 14 ust. 1 (reuse "bezwarunkowo, z wyjątkiem przypadków określonych w ustawie"), art. 15 (conditions the entity may set). Art. 6 ust. 2 limits reuse for the privacy of natural persons, including personal data. Open points: art. 5 pkt 1 of the Open Data Act speaks of information in the obliged entity's own BIP or system, whereas TED is an EU system; about 20% of documents (3,110 of 15,544, heuristic on the `author` field) have an author with a company form, and not every company is a "podmiot zobowiązany" under art. 3; treating a tender notice as an "urzędowy dokument" is a reading, not a settled point. An earlier version of this section dated the Decision 21 December 2011 and called it "the legal instrument behind that grant"; both statements were wrong and are removed. Exceptions to the grant (third-party works, identifiable private individuals) are handled in "Legal-exception and name review" below. ## What the texts are - **Genre**: public-procurement notices - tender titles, object descriptions, lots, technical specifications, selection and award criteria, deadlines, review-appeal instructions (often quoting the Public Procurement Law). - **Register**: administrative/technical Polish - dense formal prose, numbered clauses, product and standard designations, statutory citations. Contracting authority is the `author` field (100% coverage). Not OCR'd (rebuilt from structured API fields), not machine-translated. - **Time span**: publication dates `2023-02-15` - `2024-05-15`. ## Filtering (gates and counts) All steps in `src/build_ted_pl.py` against the pinned acquisition: - Acquire: 20,588 official TED Search API records (Polish-labelled notices); 1,255 via official XML fallback. Any record with a U+FFFD replacement character after cleaning is rejected. - Text: only the narrative fields (`*-proc`, `*-lot`, `*-part`, `*-glo` title/description/justification families); structured `Organization`, `Company`, `PartyName`, `PostalAddress`, `Contact`, `PartyLegalEntity`, `Person` blocks are excluded by construction. - Boilerplate: chunks whose normalized form occurs in >= max(100, 1% of documents) are dropped (one chunk, "Zamawiający nie stawia szczególnych wymagań ..."; the v0.2.6 rebuild over 15,544 documents uses a threshold of 155 documents instead of 205 and removes the same chunk). - Gates: contracting authority present; text >= 300 characters after cleaning; langid Polish (7-language set pl/en/de/cs/sk/uk/ru on the first 20,000 chars, rejection at confidence >= 0.98); exact dedup on normalized text; near dedup (5-word shingles, MinHash LSH 128 perms / 0.86 candidates, exact Jaccard >= 0.9). - Dropped: **5,044 of 20,588** - exact normalized duplicates 2,350, near duplicates 1,680, too short after cleaning 1,014. (U+FFFD, missing authority and non-Polish gates each dropped 0 in this run.) Kept: 15,544. ## PII scrubber and spot check Scrubber that ran: **pattern redaction in `src/build_ted_pl.py`** - e-mail -> `[PII]`, phone -> `[Telefon]`, labelled PESEL/NIP/REGON -> `[PII]`. This is **not** [SlayerLab/NERGAL](https://huggingface.co/SlayerLab/NERGAL); the shard must be re-run through NERGAL before it is admitted to a release (v2#60). Merge with `release: None` is fine meanwhile. **v0.2.6 rebuild.** The text was rebuilt from the official TED records, not patched: `python src/build_ted_pl.py --output --data data/ted_pl rebuild` re-fetches exactly the 15,544 shipped notice ids (Search API in batches of 100, official XML where the API has no narrative), applies `redact_pii` once to the raw text and installs the Parquet, attribution hashes, sample, stats and QA sidecars (tests: `src/test_ted_pl_contract.py`). The raw downloads are not committed. All 15,544 notices resolved, their raw bytes hash identically to the original acquisition, and no API key, terms prompt or rate limit was met. The document set and order are unchanged; nothing was added, dropped or re-deduplicated. An intermediate v0.2.6 rescrub of the shipped text (commit `1ceac95`) was replaced by this rebuild and its code removed; the extended e-mail and phone patterns it introduced are the ones applied now. Phone labels ("Tel.:", "telefon") stay in the text and only the number is replaced. | | v0.2.5 build | rescrub (`1ceac95`) | rebuild | |---|---|---|---| | Documents | 15,544 | 15,544 | 15,544 | | Characters | 112,883,460 | 112,878,041 | 112,881,592 | | Tokens | 42,475,098 | 42,471,550 | 42,473,050 | | `[Telefon]` spans / documents | 470 / 237 | 950 / 483 | 785 / 408 | | `[PII]` spans / documents | 1,080 / 703 | 1,093 / 714 | 1,093 / 714 | Against the rescrub, 164 `[Telefon]` spans in 81 documents are restored: the old phone rule had replaced money amounts, quantities, part and catalogue numbers, one procurement code and one bank account. `[PII]` does not change because labelled PESEL/NIP/REGON are still replaced (see below). A further 295 documents differ only because the phone label is now kept. One bare "Tel.:" line (ted_pl_663681-2023) is now dropped as a chunk shorter than 20 characters after redaction. **Hand-check of redactions** (false positives). Ten restored spans were read in context. All ten are correct restorations and none identifies a person: amounts ("w kwocie 199 998 723,10 zł", ted_pl_96037-2024; "520.000.000,00 złotych", ted_pl_244198-2024; "150.000.000,00 zł", ted_pl_251403-2024; "200 000 000,00 PLN", ted_pl_679171-2023), a quantity ("230 000 000 j. m.", ted_pl_141676-2024), part and catalogue numbers ("rys. nr 20.008.150.002", ted_pl_131542-2024; "150-022-002", ted_pl_236159-2024; "NR KAT. 433.302.002.0", ted_pl_232043-2024), a procurement code ("PUZ-2380-025-060-025/2024/DZ", ted_pl_220094-2024) and an amount without a currency sign ("3 000 000 000 (trzy miliardy) PLN", ted_pl_507970-2023). No NIP/REGON/KRS was restored, because their pattern did not change. A fresh sample of 12 random `[Telefon]` spans in the rebuilt text (seed 20261006) holds 11 real phone or fax numbers and 1 part number ("( 008.003.000 )", ted_pl_283348-2024). Dotted 3-3-3 part numbers are still read as phones: 5 spans in 2 documents (ted_pl_232043-2024, ted_pl_283348-2024). CPV codes are untouched. **Company and sole-trader identifiers.** Labelled PESEL/NIP/REGON numbers are replaced by `[PII]` whether they belong to a person or a company (the pattern is unchanged since v0.2.5): 242 matches in the raw text, NIP in 66 documents, REGON in 38, PESEL in none. Contractor names next to such a number were reviewed: one contractor looks like a natural person trading under their own name (ted_pl_167339-2024: "PARTNER D*** A***, (ul. P*** X, XX-XXX W***, NIP XXX-XXX-XX-XX)", masked); three more carry personal names but are partnerships (ted_pl_91828-2024, ted_pl_156428-2024, ted_pl_183186-2024). No contracting authority is a natural person. In the text these numbers are `[PII]`; the names are not redacted (NERGAL pass). Spot check of the earlier build (60 of 15,544 records, seed 20260928) found one residual contact, `ted_pl_188212-2024` ("skontaktować się z A.Decem - 815384157"): an unspaced 9-digit phone after a name. The rebuild redacts it. Other hits were benign: institutional delivery addresses (52), statutory mentions of PESEL/NIP (3). Known residual: bank account numbers (about 2,100 numbers in 1,111 documents, nearly all contracting-authority "wadium" accounts) and KRS numbers are not touched; phones written without a label and without spaces, helpline numbers shorter than 7 digits ("116 123") and role mailboxes with a missing domain ("iod@ .pl") remain; a few labelled numbers leave a trailing digit ("Tel.: 22 1551 626 5" -> "Tel.: [Telefon] 5"). ## Legal-exception and name review (scoped) Personal names are **not** redacted in this release candidate; the NERGAL pass (v2#60) handles them. A heuristic scan counts names only: 172 documents contain a candidate full name of a contact person (116 with a nominative given name from a 108-name list followed by a capitalised surname, 60 with "p."/"Pan"/"Pani" + two capitalised words, 4 overlap). It is a lower bound: it misses inflected names and unlisted given names, and it has false positives. Third-party and name cues were sampled on 60 records of the earlier build (per-notice coverage remains a release gate): free-text name cues were statutory phrasing only ("imię i nazwisko" in the Public Procurement Law's description of an appeal); third-party markers were product names (Windows Server, Microsoft 365), standard *numbers* (ISO 9001, PN-EN 14065) and legal citations/titles - no reproduced third-party texts found. ## Overlap with uzp_orzeczenia_pl and inclusion argument `uzp_orzeczenia_pl` covers the same policy domain but a different document class: appellate rulings on public procurement (KIO/SA/SN), ex-post. This slice is ex-ante tender notices. Measured text-level overlap (normalized exact-line/paragraph containment, >=30/100 chars, verbatim only): **1,082 of 191,353 lines (94,840 chars = 0.085% of this slice's chars), 0 paragraphs**; reverse direction 0.01% of `uzp_orzeczenia_pl` chars. The shared lines are statutory boilerplate (procurement-law citations common to both registers). Net: the slices are textually near-disjoint and complementary in the procurement lifecycle; ordinary cross-source dedup at integration is enough. ## Limitations Free-text personal names are not de-identified (NERGAL pass pending, v2#60). The official reuse notice has exceptions for third-party works and identifiable private individuals, so per-notice/legal review remains appropriate, and the Polish public-body basis is an inference that no lawyer has confirmed. Labelled PESEL/NIP/REGON are replaced by `[PII]` whether they belong to a person or a company, and a few dotted part numbers are still read as phones. Residual unlabelled numbers, bank accounts and role mailboxes with a missing domain are listed above. Cross-source DynaWord deduplication, benchmark-overlap checks and controlled training ablations remain pending. This administrative/technical register should be capped deliberately in a training mix.