artifacts/source_findings.md

# Source investigation findings

Running log of sources that were investigated, whether or not they were added.
**Append-only, version-independent** — unlike the `*_v0_3.md` artifacts, which are
point-in-time snapshots of one release's candidate review.

Purpose: stop the same source being re-investigated from scratch every few months.
A negative finding is as valuable as a positive one, as long as it records *what
was checked* and *what would change the answer*.

Related, don't duplicate:
- `source_scouting_v0_3.md` — v0.3 candidate shortlist (HuggingFace-centric)
- `source_candidate_audit_v0_3.md` — machine-generated audit of those HF cards
- `source_license_review_v0_3.md` — v0.3 license decision table

## How to add an entry

One `##` section per source. Keep these fields:

- **Investigated** — date, so a stale finding is visible as stale.
- **Verdict** — `added` / `blocked` / `hold` / `rejected`, plus the one-line reason.
- **Enumeration** — how you would actually get the data, or why you can't.
- **License evidence** — what you checked, with URLs. Separate what you *verified
  yourself* from what you read second-hand; label the weak ones.
- **Blockers** — what must be resolved before this can ship.
- **Lesson** — anything generalizable to other sources. Optional.

Write down the API endpoints and quirks you had to dig for. That digging is the
expensive part, and it is invisible in a verdict line.

---

## zpe.gov.pl — Zintegrowana Platforma Edukacyjna

**Investigated:** 2026-08-27
**Verdict:** `hold` — technically very tractable, blocked on per-material license
evidence. Do not build a fetcher until the license question below is answered.

Ministry of Education platform (successor to `epodreczniki.pl`), e-materials
across all school subjects. Attractive because the corpus has **no modern
instructional/pedagogic register at all**, and this is contemporary, clean,
editorially reviewed Polish prose.

### Enumeration — solved

There is no sitemap (`/sitemap.xml` → 404) and site search is client-side, but
`https://zpe.gov.pl/config.js` exposes an undocumented JSON API:

```
window.CP_CONFIGURATION = {... "apiUrl":"https://zpe.gov.pl/api/v1/" ...}
```

```
GET https://zpe.gov.pl/api/v1/search?query=&limit=20&offset=N
→ {"count": 23843, "offset": 0, "limit": 20,
   "data": [{"id": "P18l4AKOi", "title": "Fotosynteza",
             "subject": [...], "stage": [...], "tags": [], ...}]}
```

Verified directly (2026-08-27):

- `count` = **23,843** materials for the empty query.
- Deep paging works — `offset=5000` returns rows, no cursor needed.
- `robots.txt` is `User-agent: * / Allow: /`.
- Article pages `/a/<slug>/<CODE>` are **fully server-rendered**; text lives in
  `data-editor-content` blocks. No JS execution required to extract prose.
- Book pages `/b/<slug>/<CODE>` render their material list client-side (0
  server-side `/a/` links), so `/b/` is useless for crawling. Use the API.

Other endpoints probed and 404: `api/v1/materials`, `api/v1/document(s)/<CODE>`,
`api/v1/footer`, `api/v1/menu`, `api/v1/search/materials`. Only
`api/v1/search?query=` responded.

### License evidence — the blocker

**Verified directly:**

- **No document-level license marker exists in the served HTML.** The only
  machine-readable license strings are per-figure, inside
  `<div class="figure_caption_bibliography">`, e.g.
  `Źródło: GroMar, licencja: CC BY 3.0.` and
  `Źródło: Sławomir Szaruga, licencja: CC BY 3.0.` These attach to
  **illustrations, not to the prose**.
- **The site footer is `<div id="footer_app"></div>` — JS-rendered.** This
  matters: the accepted legal basis for both `govpl` and `samorzad_gov_pl` in
  `src/sources.py` is a *server-side footer string* declaring CC BY-SA 4.0.
  ZPE has no such string. **That precedent does not transfer to ZPE.**
- The platform `regulamin` states:
  *"Korzystanie z Materiału odbywać się może tylko i wyłącznie w zakresie
  dozwolonym przepisami ustawy o prawie autorskim i prawach pokrewnych"* —
  statutory permitted-use only, which is not an open license. Reading the
  surrounding clauses, this section appears scoped to material *uploaded by
  users to* the platform rather than to ministry e-materials, but the wording is
  not self-limiting and a reviewer will land on it.

**Second-hand, not verified — confirm before relying on any of it:**

- Ministry-produced e-materials are said to be CC BY 3.0, funded under
  POKL 2007–2013 / POWER 2014–2020.
- User-contributed materials are reported to carry **CC BY-NC-SA 4.0**. NC is
  already on the reject list (`source_scouting_v0_3.md`), so those must be
  excluded, not merely flagged.
- Literary works embedded in e-materials are reported to sit under a
  non-free *"licencja edukacyjna"* where authors or heirs did not agree to open
  release. This is the highest-risk content class on the platform.

### Consequences for design

- ZPE cannot be a single-license shard. It has to be **per-record license-gated**,
  `license_spdx: MIXED`, same shape as `biblioteka_nauki` and `europeana`.
- Available heuristic, **not** license evidence: ministry materials come back with
  `subject` and `stage` populated; user uploads return `"subject": [], "stage": []`
  (the first empty-query result is an all-caps user upload with both empty). Usable
  as a pre-filter, useless as provenance.
- Rough size if ~10–15k ministry materials survive the gate: **~30–100M tokens**.
  Estimate only — nothing was counted. Good register, modest scale.

### Extraction hazards (seen in a single sampled page)

- Interactive-exercise chrome: `Pokaż podpowiedź`, `Pokaż odpowiedź` buttons.
- JSON blobs embedded in `data-editor-params` attributes (`gameTitle`,
  `feedbackPositive`, …) — must not leak into `text`.
- Per-figure attribution lines repeated throughout the body.
- Glossary / `Zapamiętaj!` boilerplate repeated across many materials — needs
  frequency-based boilerplate detection, not a hand-written blocklist.

### What would unblock this

Any one of: a resolvable per-material license field (API or page markup); an
authoritative ORE/MEN page granting open reuse of e-material *text*; or written
permission. Absent that, ZPE stays out — the datasheet could not honestly answer
"where does the license statement live", which is a required field for any PR.

### Lesson

Check for a **server-rendered** license statement early. A JS-rendered footer is
not usable license evidence, and "it's a `gov.pl` domain" is not a license — the
`govpl` and `samorzad_gov_pl` entries hold up because a specific footer string
was fetched and quoted, not because of the domain.

---

## ParlaMint-PL — parliamentary debates 2015–2022

**Investigated:** 2026-08-27
**Verdict:** `ready to build, low yield` — license and distribution verified, the
fetcher is the only missing piece. But it is a **date-hole patch worth ~36M
net-new tokens**, not a diversity win. Size it before prioritising it.

Already registered in `src/sources.py` as `parlamint_pl`, but there is **no shard
in `data/`**.

### Size — measured, not guessed

From the official stats at
`github.com/clarin-eric/ParlaMint/tree/v5.0/Build/Metadata`
(`ParlaMint-overview-stats.tsv`, `ParlaMint-speeches-stats.tsv`):

```
ID  Lang  Bodies   Ts  From     To       Yrs  Mw/Yr  Mw
PL  pl    low+upp  4   2015-11  2022-06  6.7  5.35   36.06
```

- **36.06M words**, **228,326 speeches**, 4 terms, both chambers.
- Token conversion measured on our own `parliamentary` shard (same register, same
  cl100k encoder; 6,000 docs / 9.7M words): **2.654 tokens/word**, 7.24 chars/word.
- → **~96M tokens gross**, ≈1% of the current 9.68B corpus.
- Download: `ParlaMint-PL.tgz`, 185.3 MB.

### Overlap with the existing `parliamentary` shard — the real issue

ParlaMint-PL is **derived from PPC**, which is exactly what the `parliamentary`
shard is (1991–2019). Time overlap runs 2015-11 → 2019-12, about 4.2 of the 6.7
years, ≈22M words likely already present. **Net new ≈ 14M words ≈ 36M tokens**,
covering 2020-01 → 2022-06.

That estimate is soft in both directions:

- `parliamentary` has **no per-document dates** — `created` is the single literal
  `'1991-01-01, 2019-12-31'` on all 324,622 rows. The overlap cannot be measured
  by filtering, only by comparing content.
- PPC includes committee sittings and interpellations; ParlaMint-PL is plenary
  only. So ParlaMint's 2015–2019 portion is plausibly a *subset* of what already
  ships, making the real overlap higher than the time-share implies.

**Exact dedup will not catch this.** `build_dynaword.py` dedups on `sha1` of the
whole document. ParlaMint ships one document per session; PPC's segmentation
differs, so byte-identical documents are unlikely and the duplicates pass
straight through the gate. This needs **near-dedup, for which the repo has no
committed tooling** — the HPLT expansion's near-dedup was run outside `src/`.
Budget for that, or the shard silently duplicates existing data.

### Version note

**ParlaMint 5.0 does not extend Polish** — PL still ends 2022-06, same as the 3.0
deposit already cited in `sources.py` (handle `11356/1486`; 5.0 is `11356/2004`).
Upgrading buys corrections, not coverage. Neither version closes the gap to
`sejm_api` (2023 onward): mid-2022 through end-2022 stays empty.

### Verified

- CLARIN.SI handle `11356/1486` resolves (HTTP 200). The `upstream` URL recorded
  in `sources.py` is correct and live.
- License: **CC BY 4.0**, stated on the repository record.
- ParlaMint 3.0 overall: 26 comparable corpora, 7,468,674 utterances,
  1,147,800,265 words; debates mostly 2015 → mid-2022.
- The distribution ships a **plain-text (`.txt`) edition** alongside TEI/XML and
  linguistically-annotated `.ana.xml`. Filenames are per-session, e.g.
  `ParlaMint-PL_2017-07-20-sejm-46-3.txt`. Use the `.txt` edition — no TEI parsing
  needed.
- Polish covers **both chambers**: `-sejm-` and `-senat-` session files.

### Why it is still worth doing

Fills 2020 → mid-2022, between `parliamentary` (ends 2019) and `sejm_api` (2023
onward). Same CC-BY-4.0 family as existing sources, so there is no new license
question to answer, and the plain-text edition means no TEI parsing. Cheap. Just
do not expect it to move the totals — and do not build it before a source that
adds a register the corpus lacks.

### Drift found while checking

`src/sources.py` (the `parlamint_pl` `provenance` field) states the corpus was
*"Fetched directly from the CLARIN.SI plain-text edition (ParlaMint-PL.tgz) by
`src/fetch_parlamint.py`"*. **That script does not exist in `src/`.** The
provenance describes a fetch that no committed code performs. Either write the
script or correct the field before this ships.

**Resolved 2026-10-05:** the `parlamint_pl` entry was removed from `src/sources.py`
(no fetcher, no data, never released). To add ParlaMint-PL later, write the fetcher and
register it again as a new source.

---

## Sejm API — interpelacje and zapytania poselskie

**Investigated:** 2026-08-27
**Verdict:** `hold — recheck`. Not a license problem; `api.sejm.gov.pl` was
unreachable from the investigating network on the day, so the endpoints could not
be confirmed. Retry before writing anything off.

### What happened

- `api.sejm.gov.pl` resolves (`194.41.12.231`) but every request hangs and then
  fails — plain `curl` times out at 45s, a proxied fetch returns `ECONNRESET`.
- `https://www.sejm.gov.pl/` responds normally (302), and other external hosts
  were fine the same session, so this is specific to the API host: either an
  outage, or that host filtering the source network.
- `src/fetch_sejm_api.py` (committed, working) uses this same host, so the API
  is normally reachable. Treat as transient until proven otherwise.

### Unverified — what to check on retry

- Whether `sejm/term{N}/interpellations` and `sejm/term{N}/writtenQuestions`
  exist, and their paging parameters.
- Whether the **ministry's reply body** is retrievable as text, or only as a
  linked PDF. This decides whether the source is worth building: the replies are
  the substantial half.
- Document counts per term.

### Why it is on the list

Reuses the existing `fetch_sejm_api.py` legal basis unchanged — official
parliamentary materials outside copyright under art. 4(2) pr. aut., reusable
under the Open Data Act. Question/answer prose is argumentative register, which
the corpus is short on. No new license work needed.

---

## CLARIN-PL DSpace (PELCRA Spokes, and the `1000_novels` upstream)

**Investigated:** 2026-08-27
**Verdict:** `blocked — temporary`. The repository is down for reorganization.
Nothing to decide until it returns.

### Verified

- `https://clarin-pl.eu/dspace/` serves a maintenance notice:
  *"Trwają prace nad repozytorium. Prowadzimy prace serwisowe oraz reorganizację
  repozytorium DSpace… Repozytorium będzie ponownie dostępne wkrótce."*
- Every item handle under it 404s: `11321/312`, `11321/544`, `11321/545`,
  `11321/547` — nginx 404, not a DSpace "item not found".
- The global resolver is broken for this prefix too: `hdl.handle.net/11321/*`
  returns **HTTP 500**.

So no PELCRA record (Spokes, EMO, PARL, EMI) can have its license checked right
now. Conversational Polish remains the corpus's biggest register gap, so this is
worth re-checking periodically.

### Side effect on shipped data

`src/sources.py` records `https://clarin-pl.eu/dspace/handle/11321/312` as the
`upstream` for **`1000_novels`**, which is a released shard. That URL currently
does not resolve. Presumed temporary, but the citation in a published datasheet
is dead as of this date.

### Expected concern when it returns

PELCRA spoken corpora are transcribed private conversation. Expect NC-flavoured
or restricted licensing, and expect a personal-data question — these recordings
involve identifiable private individuals, not public officials acting officially.
Both must be resolved before use.

---

## NIK — Najwyższa Izba Kontroli audit reports

**Investigated:** 2026-08-27
**Verdict:** `hold` — permissive but **non-standard** license, and enumeration is
unsolved. Viable with work; not a quick win.

### License evidence — verified

`https://www.nik.gov.pl/zasady/prawa-autorskie.html` states:

> *"Korzystanie z materiałów publikowanych w serwisach internetowych Najwyższej
> Izby Kontroli nie wymaga uzyskania zgody NIK. Nie dotyczy to elementów odrębnie
> zastrzeżonych (do których prawa autorskie przysługują osobom trzecim), w
> szczególności oznaczonych materiałów zdjęciowych oraz tekstów publikowanych
> pierwotnie w innych źródłach."*

Attribution to *"Najwyższa Izba Kontroli (nik.gov.pl)"* is requested, not required.

Assessment: a **custom permissive notice, not a standard open license** — no CC,
no CC0, no SPDX identifier. It carries a third-party carve-out that matters here,
because audit reports quote external sources. A separate and probably stronger
argument is that audit reports are *materiały urzędowe* under art. 4 pkt 2 pr.
aut., the same basis as `dziennik_ustaw` and `sejm_api`. If this source is built,
lead with the statutory argument and cite the site notice as corroboration.

### Enumeration — unsolved

- `robots.txt` allows crawling, but disallows `/szukaj/`, `/p/`, `/g2/`,
  `/cms/`, `/do_druku/`. `/g2/` looks like the file store, which likely means
  **report PDFs are off-limits to crawlers**.
- `/kontrole/wyniki-kontroli-nik/` returns 200 (~72 kB) but the report list is
  **JS-rendered**: zero report links in the server HTML, only pagination
  scaffolding (`fmax,js,strona,1.html`).
- The `/prosta/` ("simple") variant is also JS-rendered — 0 report links.
- No `/sitemap.xml` (404).

Same shape as ZPE: a backend API almost certainly exists and would need to be
found (check the page's JS config for an `apiUrl`, as worked for ZPE).

### Verdict rationale

Register value is real — long-form analytical prose, unlike statute text. But
between a bespoke license, a third-party-content carve-out, PDF paths blocked by
`robots.txt`, and unsolved enumeration, this is several days of work. Lower
priority than ParlaMint-PL or Sejm interpelacje.

---

## Billion-scale survey — what can actually move the totals

**Investigated:** 2026-08-27

Context: most candidates land at 30–200M tokens, which is noise against a 9.68B
corpus. This entry records the only options measured at **500M–5B**, i.e.
comparable to `european_hplt_v3_pl` (2.56B), `eurlex` (2.38B), `parliamentary`
(1.65B). All numbers below were measured on the day, not taken from cards.

Token conversion throughout: **2.654 tokens/word**, measured on the repo's own
`parliamentary` shard (6,000 docs / 9.7M words, cl100k). Use it instead of
guessing — Polish runs far above the ~1.3 rule of thumb for English.

| candidate | measured | est. tokens | license | blocker |
|---|---|---|---|---|
| SAOS judgments | 540,338 docs, mean 3,067 w | **~4.4B** (floor ~2.8B) | art. 4 pkt 2 pr. aut. | register: pure legal |
| FineWeb-2 `pol_Latn` | 152,003,298 docs, 225.9 GB | ~83B raw, **4–8B** filtered | ODC-By | dedup vs existing HPLT |
| PleIAs/Polish-PD | 68,176 docs, 0.75B words | **~2.0B** nominal | public domain | OCR corruption |
| HPLT leftovers | shards `8_5`, `8_6`, 31.6 GB | ~230M | CC0 | none — just small |

### SAOS — biggest clean win, already half-built

`https://www.saos.org.pl/api/search/judgments?pageSize=1&pageNumber=0` reports
`totalResults: 540338`. Sampled 100 judgments from
`api/dump/judgments?pageSize=100&pageNumber=0`: **mean 3,067 words, median 1,920,
p10 816, p90 7,197**, court types `COMMON` and `CONSTITUTIONAL_TRIBUNAL`.

540,338 × 3,067 = 1.66B words → **~4.4B tokens**. Using the median instead, to
guard against long-tail skew, still gives ~2.75B. Either way this is **larger
than any shard currently in the corpus**.

`src/fetch_saos.py` already exists (uncommitted as of this date) and documents the
legal basis: judgments are *materiały urzędowe* outside copyright under art. 4
pkt 2 pr. aut., redistributed by SAOS (CeON/ICM UW + Fundacja ePaństwo) through a
bulk-dump API built for exactly this.

Open question before building: judgments are court-anonymised, but that should be
**verified on sampled text**, not assumed — this is the one candidate here where
personal data is a live risk rather than a formality.

Cost of shipping it: legal/official goes from **48% to roughly 65%** of the
corpus. That is the trade, and it should be a deliberate decision rather than a
side effect of it being the easy build.

### FineWeb-2 Polish — only billion-scale contemporary prose

`datasets-server` size endpoint, config `pol_Latn`: **152,003,298 rows /
225.9 GB**. At the measured 0.367 tokens/char that is **~83B tokens raw**.

License `odc-by`. Same argument shape as the accepted HPLT basis: the packaging
is licensed, the underlying web text is not owned by the publisher. Note that
ODC-By is **attribution-required**, unlike HPLT's CC0 — a different obligation
shape, and it has to be stated in the datasheet.

#### Dedup is not the bottleneck — this was checked, not assumed

The obvious worry is overlap with the existing HPLT shard, since both derive from
Common Crawl. It is the *smallest* of the problems:

- The existing HPLT shard is **1.6M docs against FineWeb-2's 152M**. Even total
  overlap removes ~1% of candidates (~12% if working from the HQ subset below).
- **Internal dedup is already done upstream.** The schema carries
  `minhash_cluster_size`, i.e. FineWeb-2 ran MinHash per language before release.

Cross-dedup against HPLT is still worth doing for hygiene, and the repo still has
no committed near-dedup tooling — but it is not what makes this source expensive.

#### The actual blocker: the HPLT gate cannot be ported

FineWeb-2 `pol_Latn` columns are:

```
text, id, dump, url, date, file_path, language,
language_score, language_script, minhash_cluster_size, top_langs
```

`src/clean_hplt_v3.py` filters on **WDS quality score, register class
(ID/OP/HI/IN/NA), and MT-probability**. *None of those exist in FineWeb-2.* Every
signal the existing web pipeline depends on is absent, so "filter it like we
filtered HPLT" is not an option — it would mean building quality classification
over 152M documents from scratch.

#### `epfml/FineWeb2-HQ` — the better target

A published quality-filtered derivative, config `pol_Latn`:

- **13,384,885 rows**, `odc-by`, 180.4 GB parquet (size dominated by an
  `embeddings` column, not text).
- Adds a **`quality_score`** column, retained per document, so the selection
  threshold stays a tunable knob instead of a baked-in decision — and the
  classifier is someone else's published work, which is easier to cite than a
  home-grown one.
- Sampled 880 docs across 22 evenly-spread offsets: **mean 2,663 chars/doc**
  → ~13.1B tokens for the whole HQ subset. Larger than the entire current corpus.

#### quality_score → token yield

Distribution from the `statistics` endpoint (computed on a **partial 407,184-row
sample**, not all 13.4M): min 0.013, max 0.9999, mean 0.145, **median 0.0496** —
heavily right-skewed, so most of HQ still scores low.

Doc fractions from that histogram, mean lengths from the 880-doc sample:

| `quality_score` ≥ | % docs | docs | mean chars | est. tokens |
|---|---:|---:|---:|---:|
| 0.00 (all) | 100% | 13,384,885 | 2,663 | **13.1B** |
| 0.11 | 31.0% | 4,149,000 | 2,375 | **3.6B** |
| 0.21 | 20.0% | 2,677,000 | 2,074 | **2.0B** |
| 0.31 | 14.6% | 1,954,000 | 1,929 | **1.4B** |
| 0.41 | 11.1% | 1,486,000 | 1,912 | **1.0B** |
| 0.51 | 8.5% | 1,138,000 | 1,798 | **0.75B** |
| 0.70 | 4.8% | 642,000 | 1,750 | **0.41B** |
| 0.90 | 1.8% | 241,000 | 914 | **0.08B** |

A second pass computing fractions directly from the 880-doc sample agreed with
these within ~15%. Treat them as ±20%, good enough to pick a threshold.

**Watch out: `quality_score` anti-correlates with document length.** Mean length
falls from 2,663 chars at no threshold to 914 at ≥0.90. The classifier is not
rewarding long-form prose, so raising the threshold costs more tokens than the
doc-count alone implies, and it may actively select *against* the extended prose
this corpus is short on. Sample the text at whatever threshold is chosen before
committing to it.

#### Recommended operating point

**Threshold ≈ 0.20 → ~2.7M docs, ~2.0–2.3B tokens.** Comparable to the largest
existing shards (`european_hplt_v3_pl` 2.56B, `eurlex` 2.38B) and it *improves*
the balance rather than wrecking it: web goes 26% → ~41%, legal drops 48% → ~39%.

Taking all 13.1B instead would put the corpus at ~22.7B with **web at ~69%** —
trading "two-thirds legalese" for "two-thirds crawl". Same failure, different axis.

#### Work order if this is built

1. Pick HQ vs full, and a `quality_score` threshold (see above).
2. **Build the PII scrub.** `sources.py` already claims one for the HPLT shard
   that has no implementing code in `src/`; shipping a second web source with the
   same gap is indefensible.
3. Near-dedup against the existing HPLT shard — needed, but ~1–12% effect.
4. Compute. Tokenizing billions of tokens is the genuinely expensive step.

### PleIAs/Polish-PD — good on paper, rough in practice

68,176 rows. Sampled `word_count` at four offsets (n=200): **mean 10,971 words,
median 8,468** → ~0.75B words → **~2.0B tokens**. Columns are `identifier`,
`file`, `title`, `creator`, `language`, `word_count`, `character_count`,
`complete_text`. Public domain, pre-1884 monographs and periodicals, largely
sourced from Polish digital libraries (`jbc.bj.uj.edu.pl` in the sample).

Register-wise this is the most attractive of the three — books and periodicals,
against a corpus that is only 1.9% literature.

But the OCR is bad. First sampled record (an 1817 newspaper) opens:

```
U
RUSKI
czwaRTEŁ SONAER
5 PC Nę R
WIADOMOŚC
```

with pre-1884 orthography throughout (`Nayiaśnieyszy`, `roskazać`, `klassy`).

**The existing OCR gate will not catch this.** `build_dynaword.py` applies
`MIN_ALPHA_RATIO = 0.70` when `is_ocr: True`, which tests the ratio of letters to
symbols. Garbled *words* made of ordinary letters pass it cleanly. Filtering this
source needs a dictionary- or perplexity-based check, which does not exist here.
Treat the 2.0B as nominal; real usable yield is materially lower.

### HPLT is exhausted — correction to an earlier assumption

`https://data.hplt-project.org/three/sorted/pol_Latn/` lists **exactly 19 shards**:
`10_1`, `9_1`–`9_4`, `8_1`–`8_6`, `7_1`–`7_4`, `6_1`–`6_2`, `5_1`–`5_2`.
**There are no WDS bins below 5 for Polish.**

`sources.py` records 17 of those as ingested. The only unused shards are
**`8_5` (24.24 GB)** and **`8_6` (7.37 GB)** — roughly 230M tokens at the
post-filter rate implied by the existing shard.

So "add more HPLT bins" is not a path to billions. It is one small top-up, and
then the source is finished.

### Conclusion

Every clean billion-scale Polish source is **legal, official, or archaic** —
which follows directly from what falls outside copyright. The only billion-scale
contemporary general prose is web crawl, and that is a re-derivation of Common
Crawl the corpus already draws on.

If the goal is raw scale: SAOS. If the goal is a corpus that is not two-thirds
legalese, **there is no billion-scale option**, and the honest plan is a number of
50–200M-token sources across different registers.

---

## common_voice_pl — Common Voice Polish validated sentences (v26.0)

**Investigated** — 2026-09-05 (review round 2026-09-29).

**Verdict** — `added` with a **unit caveat** (proposed with `release: None`, merge
blocked on a maintainer unit decision). 45,043 unique validated speech prompts /
995k `cl100k_base` tokens, CC0-1.0, from Mozilla Common Voice Scripted Speech
26.0 (Polish).

**Enumeration** — official dataset record on Mozilla Data Collective
(<https://mozilladatacollective.com/datasets/cmqinmu2a00winq07gyrtri0q>, id
`cmqinmu2a00winq07gyrtri0q`, release 2026-06-12), read via a pinned row mirror
with DuckDB column projection (the 4.48 GB audio is never pulled). 138,798
validated clip rows -> 45,043 unique normalized sentences (91,198 repeated
clips collapsed; 2,538 too-short and 19 too-long rows rejected).

**License evidence** — verified AT THE ORIGIN: the MDC dataset record's license
field is "Creative Commons Zero v1.0 Universal (CC0-1.0)", and the same
`license`/`license_url` ride along per source row (build rejects non-CC0 rows).
**Terms covenant**: MDC Terms of Use (Data Consumer License, clause viii)
prohibit re-identifying individual contributors/sources — carried to downstream
users in the datasheet and registry.

**Blockers** — (1) **unit decision**: every doc is a single speech prompt (min 12
/ median 56 / max 153 chars; 0 docs reach the 200-char corpus gate, no
concatenation exists) — accept sentence units with a per-source gate override +
training cap, or withdraw as unit-incompatible; (2) before release: NERGAL pass
(full-shard residual scan was clean — 0 contacts across 45,043 docs — but the
prompts are user-contributed speech material).

**Lesson** — speech corpora's natural unit (the prompt sentence) is smaller than
a pretraining corpus's document gate; measure the length distribution BEFORE
promising the source, and keep the unit question in the PR description up
front — it is the first thing a reviewer will ask.

---

## sejm_interpellations — 10th-term written Q&A (interpellations + replies)

**Investigated** — 2026-09-05 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 25,950 documents /
28.2M `cl100k_base` tokens of written MP questions and ministry replies —
argumentative administrative prose, a register the floor-speech shards do not
carry.

**Enumeration** — `GET /sejm/term10/interpellations` and `.../writtenQuestions`
plus `/reply/{key}/body` on api.sejm.gov.pl (the endpoints the 2026-08-27
"hold — recheck" entry above could not confirm; they exist and page fine).
PDF-only replies (`onlyAttachment`) and keyless stubs skipped. Term 9 returned
`X-Total-Count: 0`; terms 7–8 skipped (PPC covers through 2019).

**License evidence** — same basis as `sejm_api`: Copyright Act art. 4(2)
(<https://eli.gov.pl/api/acts/DU/2025/24/text/O/D20250024.pdf>) + Open Data Act
arts. 2(12), 5, 14, 17 (<https://eli.gov.pl/api/acts/DU/2023/1524/text.html>).
No CC grant claimed.

**Blockers** — none to merge with `release: None`. Before release: **NERGAL
pass** — interpellations name private constituents (constituent letters and
citizen cases are quoted in questions), which is exactly the case the old
`scrub_pii` (#20) could not cover and #38 moved out of this repo. Caveats for
the card: single term (10th), private citizens named.

**Lesson** — the header chrome (nr / recipient / title / signatories / date)
belongs in meta, not text: stripping it makes the body comparable across
question and reply docs, and the before/after is the fastest reviewer evidence
for any Sejm HTML source.

---

## Edukacja Medialna — media-literacy lessons (`edukacja_medialna_pl`)

**Investigated** — 2026-09-06 (review round 2026-09-28).

**Verdict** — `added`. 252 records / 728,987 `cl100k_base` tokens of CC BY-SA
3.0 lesson materials — small, but a school-teaching register nothing else in the
corpus covers.

**Enumeration** — <https://edukacjamedialna.edu.pl/lekcje/> lists every lesson
(255 at fetch). Each lesson page carries its source XML: a `utwor` document with
Dublin Core fields and the authored `powiesc` body. `src/build_edukacja_medialna_pl.py`
fetches page + XML, checksums both (`page_sha256`, `xml_sha256`) and builds from
the pinned `artifacts/raw_xml.jsonl.gz`; nothing else is needed.

**License evidence** — verified at the origin (the mirror is not evidence for
itself): every lesson page ends with `Licencja: Creative Commons Uznanie
autorstwa - Na tych samych warunkach 3.0` linking
<https://creativecommons.org/licenses/by-sa/3.0/> (e.g.
<https://edukacjamedialna.edu.pl/lekcje/abc-bezpieczenstwa-w-sieci/>), the site
footer repeats the grant (<http://creativecommons.org/licenses/by-sa/3.0/deed.pl>),
and the source XML carries `dc:rights` + `dc:rights.license`. The build gate
requires the statement on page AND XML. Per-record `rights`/`rights_urls` and
hashes are pinned in `attribution.jsonl`.

**Blockers** — none to merge with `release: None`. Before a release: NERGAL pass
(the in-build redaction is e-mail/labelled-phone patterns only — 1 match), cross-source
exact/near dedup, benchmark-overlap check, and maintainer review of embedded
third-party excerpts and of the corpus-level BY-SA 3.0 treatment (proposal in the
datasheet: per-record attribution + share-alike note, adaptation released under
CC BY-SA 4.0).

**Lesson** — cite the license where the publisher states it (per-lesson
`Licencja:` line + XML rights fields), never a mirror. Pin the statement with
`page_sha256`/`rights_urls` so it survives site redesigns.

---

## ted_pl — TED Polish public-procurement notices

**Investigated** — 2026-09-08 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 15,544 notices / 42.5M
`cl100k_base` tokens of Polish administrative/technical procurement prose
(2023-02-15 – 2024-05-15).

**Enumeration** — 20,588 official TED Search API records for Polish-labelled
notices (1,255 via official XML fallback); the PleIAs mirror is a pinned
identifier index only (its Polish preview carries U+FFFD damage). Narrative
fields rebuilt from API, structured contact/address/person blocks excluded by
construction.

**License evidence** — TED legal notice
(<https://ted.europa.eu/en/legal-notice>): procurement notices freely reusable
commercially/non-commercially unless otherwise noted; metadata CC0 (separate
grant). Legal instrument behind it: **Commission Decision 2011/833/EU on the
reuse of Commission documents**
(<https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32011D0833>).
Text recorded as `LicenseRef-TED-Notice-Reuse`, not relabelled CC0.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass
and per-notice legal-exception review — the 60-record spot check found **one
real residual contact** (`ted_pl_188212-2024`: "A.Decem - 815384157" — name +
unlabelled phone the patterns missed) and free-text names are un-scoped beyond
the sample. Text overlap with `uzp_orzeczenia_pl` is negligible (0.085% chars,
0 paragraphs — shared statutory citations), so ordinary cross-source dedup
suffices.

**Lesson** — unlabelled phone numbers beside personal names defeat
label-anchored patterns (same lesson as `open_icm_pl`'s glyph noise): a zero
pattern count is not evidence in contact-bearing registers — run the spot check
and the model scrubber.

---

## kultura_bez_barier_pl — cultural-accessibility guides (FKzB)

**Investigated** — 2026-09-07 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 9 guides / 218k
`cl100k_base` tokens (316 pages) of prepared didactic Polish for cultural-sector
accessibility work - tiny, but a clean CC BY-SA 3.0 source with preserved raw
PDFs and per-record evidence.

**Enumeration** — the foundation's catalogue <https://kulturabezbarier.org/publikacje/>:
12 linked publications discovered, 9 retained. 3 rejected conservatively: 2 NIMOZ
co-published volumes (PDF copyright notices name the third-party institute) and
1 for legacy-font extraction damage.

**License evidence** — the catalogue's own statement (quoted in the datasheet)
offers the foundation-authored publications under CC BY-SA 3.0 with attribution
and share-alike terms; reading is catalogue-level ("plik pierwotny"), pictograms
explicitly included, no per-PDF embedded assertion claimed. Per-record
corroboration: PDF metadata + credit lines; `attribution.jsonl` pins the
evidence URL and scope note.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass —
the all-9 spot check found **real residual contacts** in
`podsumowanie_wizyt_studyjnych` (coordinator names + phones under `Tel./SMS`
label variants the pattern misses, e-mails beside them redacted) — and
maintainer sign-off on the catalogue-level scope reading.

**Lesson** — catalogue-level license statements need the reading written down
(what "je" covers), and the safest per-record confirmation is negative: reject
every item with third-party authorship signals, as this PR does with the NIMOZ
volumes. Also: `Tel./SMS`-style label variants defeat label-anchored phone
patterns — third independent pattern-evasion case in this review round.

---

## plwiki_talk — Polish Wikipedia talk pages (ns 1+5)

**Investigated** — 2026-09-11 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 43,054 pages / 55.6M
`cl100k_base` tokens of **user-generated** editorial discussion — article talk
(ns 1, 41,828 pages) and project talk (ns 5, 1,226 pages). Exact namespace
complement to `wikipedia` (ns 0) and `plwiki_user_talk` (ns 3).

**Enumeration** — the same pinned `plwiki-20260901-pages-meta-current` dump as
the ns-3 slice (7 sha1-verified shards); 544,591 non-redirect talk pages
enumerated, 43,054 retained after QA/dedup (500,652 dropped as <500 chars).

**License evidence** — Wikimedia Terms of Use section 7 (CC BY-SA 4.0):
<https://foundation.wikimedia.org/w/index.php?title=Policy:Terms_of_Use/pl&oldid=584706>.
Per-record canonical URL + page history is the attribution trail.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass
(user-generated talk; a 60-record residual spot check found only a DOI false
positive and public signatures, but free text can hide more), and text-level
cross-shard dedup per the measured overlap with `plwiki_user_talk` / `wikipedia`
(review thread).

**Lesson** — namespace partitioning prevents page-level duplication but not
text-level repetition: talk pages quote article text and share templated notices
across namespaces. Measure overlap at text level before claiming novelty — the
ID-only audit said "0 collisions" and proved nothing.

---

## plwiki_user_talk — Polish Wikipedia user talk, ns 3

**Investigated** — 2026-09-11 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 380,527 records / 381.4M
`cl100k_base` tokens of **user-generated** editor-to-editor conversation — a
register the corpus otherwise does not draw from.

**Enumeration** — official dump `plwiki-20260901-pages-meta-current`, 7
sha1-verified shards against `dumpstatus.json`; ns=3 non-redirect pages with
text = 450,867 candidates. Rebuilt deterministically by
`src/build_plwiki_usertalk_pl.py` from the pinned shards.

**License evidence** — Wikimedia Terms of Use section 7 (community content under
CC BY-SA 4.0):
<https://foundation.wikimedia.org/w/index.php?title=Policy:Terms_of_Use/pl&oldid=584706>
(verified, HTTP 200). Per-record canonical page URL + page history is the
attribution trail, pinned in `attribution.jsonl` (380,527 lines, LFS).

**Blockers** — none to merge with `release: None`. Before release: a NERGAL pass
(user-generated conversation is the policy's core case; the in-build redaction is
patterns only and a 60-record residual spot check found 0 hits but covers 0.016%),
and cross-shard cleanup against `plwiki_talk` for 47 shared long notice lines
(99.3% of the measured ns3↔ns1/5 line overlap; genuine thread copies are ~0.13% of
user-talk chars).

**Lesson** — text-level overlap audits beat ID audits: the ID comparison reported
"0 collisions", the text pass found 28.7% char containment against `plwiki_talk` —
99.3% of it just 47 template/notice lines shared across namespaces. Also:
line-level boilerplate rules need their length cap revisited (the <160-char rule
lets long templated notices survive in both corpora).

---

## open_agh_chemistry_pl — Open AGH chemistry textbooks

**Investigated** — 2026-09-05 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 370 modules / 564k
`cl100k_base` tokens from four Polish AGH chemistry textbooks (general,
inorganic, polymer, corrosion) — didactic LaTeX-bearing prose.

**Enumeration** — the AGH e-textbook platform
(<https://epodreczniki.open.agh.edu.pl/handbook/{29,1394,37,1893}>); 374
modules discovered, 370 retained (each with a selection decision). Pinned
snapshot `PiotrSty/open-agh-chemistry-pl@07d38bfd` with raw module HTML.

**License evidence** — verified AT THE ORIGIN: each book's official EPUB rights
page (served by AGH at `/rpc/preview/handbooks/{id}/epub/light`, preserved
verbatim with `rights_sha256` in `books.json`) states "Publikacja udostępniona
jest na licencji Creative Commons Uznanie Autorstwa - Na tych samych warunkach
4.0…" with attribution terms and a "Wersja oryginalna" marking requirement.
Per-book, not catalogue-level — the strong pattern.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass
(two-pattern redaction only; the 60-record spot check was clean — DOIs only),
cross-source dedup and benchmark checks.

**Lesson** — repo hygiene findings worth keeping: (1) `sources.py` line endings —
`main`'s own blob is CRLF today, so the AGENTS.md "commit LF" rule as written
would CREATE whole-file rewrites; the minimal-diff rule is "match main", and the
repo-wide normalization is maintainer work; (2) the entry shipped without
`added`/`release` — the registry contract check belongs in the contract test.

---

## openstax_pl — OpenStax Poland textbooks (8 volumes)

**Investigated** — 2026-09-05 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 1,422 pages / 4.63M
`cl100k_base` tokens of Polish academic textbook prose (physics ×3, psychology,
micro/macro, marketing, nutrition) — **human translations of the English
OpenStax originals** by named academics.

**Enumeration** — openstax.org book containers (Rex/Connexions HTML) via
openstax.pl catalog; 1,612 pages discovered, 1,422 retained (each with a
decision; MathML-conversion failures rejected). Pinned snapshot
`PiotrSty/openstax-pl-textbooks@5b31f74e`.

**License evidence** — verified AT THE ORIGIN per book: each Polish edition's
foreword page (openstax.org/books/<slug>/pages/przedmowa) states the CC BY 4.0
grant with the attribution condition "uznania autorstwa OpenStax i osób
podpisanych pod oryginałem"; links per book in the datasheet/PR, excerpts +
sha256 preserved in `books.json`.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass
(two-pattern redaction only; spot check of 60 was fully clean), cross-source
dedup and benchmark checks (exercises/tables may reference omitted figures).

**Lesson** — for translated editions, "by whom" is a licensable-quality question:
the foreword credits (named translators per chapter) are the evidence — pin
them per record (`edition_contributors`) instead of asserting "translated" in
the abstract. Same for MT: state its absence explicitly when the credits are
human.

---

## wiadomosci_statystyczne_pl — Wiadomości Statystyczne / The Polish Statistician (GUS)

**Investigated** — 2026-09-10 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 169 articles / 1.5M
`cl100k_base` tokens of peer-reviewed statistics-journal content (research,
statistics history, conference reports, obituaries) — small, but a formal
academic register with a hard per-article license gate.

**Enumeration** — deterministic crawl of the server-rendered
`https://ws.stat.gov.pl/` `/Archives` index: 460 articles (2022+), 315 with the
explicit per-article CC BY-SA 4.0 line, 194 acquired, 182 after QA/dedup, minus
13 exact title/DOI matches against `biblioteka_nauki` -> 169.

**License evidence** — GUS journals under CC BY-SA 4.0 since 2022-01-01
(<https://nauka.stat.gov.pl/News/Info/60>) AND a verbatim per-article
"udostępniony na licencji CC BY-SA 4.0" line on each landing page (preserved
HTML snapshots in the source repo; the line also appears in the PDF text).
Pre-2022 articles show only the site-wide footer and are excluded — that
archive's license status is unresolved.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass
(pattern redaction covers e-mail/labelled-phone only; a 60-record spot check
found just DOIs/ORCIDs, but case-study text can hide more) and cross-source text
dedup (the 13 exclusions were metadata-level title/DOI matches only).

**Lesson** — the per-article license line kept verbatim in the record text makes
the grant self-evidencing in the corpus itself — worth copying for other
publishers that stamp each article.

---

## sejm_committee_transcripts — Sejm committee full transcripts (terms 9–10)

**Investigated** — 2026-09-18 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 653,411 speaker turns /
166.3M `cl100k_base` tokens of spoken multi-party committee discussion — a
working register the plenary-only political sources do not cover.

**Enumeration** — api.sejm.gov.pl: `GET /sejm/term{n}/committees` and
`.../sittings`, then `.../sittings/{num}/pdf` for the "pelny zapis" transcripts.
9,175 term-9/10 sittings known; 8,658 have a zapis PDF upstream, 455 have none,
62 failed after retries; 7,746 sittings survive joint-sitting dedup.

**License evidence** — official parliamentary materials: Copyright Act art. 4(2)
(<https://eli.gov.pl/api/acts/DU/2025/24/text/O/D20250024.pdf>) + Open Data/PSI
Act arts. 2(12), 5, 6, 14, 15, 17
(<https://eli.gov.pl/api/acts/DU/2023/1524/text.html>). No CC grant claimed;
evidence snapshot in `sejm_committee_transcripts.license-evidence.json`.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass
(454 pattern redactions cover only e-mail/PESEL/NIP/REGON/labelled phone/IBAN;
quoted citizens' statements remain), near-dedup vs `dziennik_ustaw`/`eurlex`
(embedded statute/print quotations), benchmark-overlap check.

**Lesson** — `pdftotext -layout` + role-anchored speaker labels reconstruct
turns well (98.9% of the ASW pilot's audited corrected characters), but it is
the pilot comparison that makes the reconstruction auditable — keep that step
for future spoken sources.

---

## Studia BAS — Kancelaria Sejmu research & public-policy journal (2021–2025)

**Investigated:** 2026-08-29
**Verdict:** `ready` — official open-access transition under **CC BY 3.0 PL** starting with issue **2(66)/2021**; verified per-article license statements and clean born-digital PDF text layer.

Quarterly peer-reviewed journal published by the **Biuro Analiz Sejmowych (BAS)** of the Chancellery of the Sejm (Kancelaria Sejmu). Focuses on public policy, socio-economic research, constitutional and EU law, public finance, and legislative commentary. Provides contemporary, formal academic/policy analysis prose — a critical register bridging official administrative/legal texts and general expository literature without being raw statutory legalese or web crawl.

### Archive enumeration & fetch route — verified

- **Archive URL:** `https://studiabas.sejm.gov.pl/journal.nsf/pubBAS.xsp?view=12&seria=4&lang=PL`
- **Enumeration:** The Lotus Domino / XPages archive portal uses dynamic AJAX updates. Querying issues by year (2021–2025) requires parsing the initial `xspForm` state (`$$viewid`, `$$ajaxid`, `cbYear`) and submitting multipart POST requests.
- **PDF Resolution:** Issue records point to download endpoints on `orka.sejm.gov.pl` (either direct issue PDF paths or the download-counting agent `LiczOpen?OpenAgent&<UNID>`).
- **WAF / Network:** `studiabas.sejm.gov.pl` is fronted by Imperva WAF, requiring browser-like TLS handshakes (`curl_cffi` impersonating Chrome) for reliable programmatic querying and retrieval.

### License evidence — verified cutoff from issue 2(66)/2021

- **Historical cutoff:** Issues 1–65 (prior to 2/2021) were published under traditional all-rights-reserved copyright.
- **Open-access grant:** Starting with issue **2(66)/2021** (issue 66), the journal officially transitioned to open access under **Creative Commons Attribution 3.0 Poland (CC BY 3.0 PL / `CC-BY-3.0-PL`)**.
- **Dual-gate verification:** Ingestion enforces a hard issue boundary (`issue >= 66`) and additionally validates the presence of the explicit CC BY 3.0 PL statement inside every article PDF (*"Artykuł został udostępniony na licencji Creative Commons – Uznanie Autorstwa 3.0 Polska (CC BY 3.0 PL)"*). Issues or articles lacking this statement fail closed.

### PDF characteristics, text extraction & non-OCR status

- **Born-digital PDFs (`is_ocr: False`):** Whole-issue PDFs feature high-fidelity digital text layers, not scanned or OCR-derived text.
- **Article segmentation:** Multi-article issue PDFs are sliced at article title pages identified via title font size (`max_size >= 12`), registered DOI (`10.31268/StudiaBAS.YYYY.NN`), and the in-PDF CC BY license block.
- **Layout normalization:**
  - Running headers (page numbering, journal title, ISSN `2082-0658`, `studiabas.sejm.gov.pl`) and author affiliation footnote blocks are stripped from the main body.
  - PDF line-break hyphenation is reconstructed for Polish compounds (e.g. `przedsię-\nbiorców`, `rolno-\n-spożywczy`).
  - Paragraph breaks are preserved and reconstructed using vertical spacing, indentation, and heading font shifts.
- **Polish scope:** Ingestion extracts the Polish title, Polish abstract (*Streszczenie*), Polish keywords (*Słowa kluczowe*), and full Polish article body. Incidental English titles/abstracts (*Abstract*, *Keywords*) in dual-language front matter are excluded from the main text.

### Observed yield (2021–2025)

- **Extracted articles:** **115 articles** across issues 66 to 81 (2021–2025).
- **Retained articles:** **112 articles** after dropping 3 entirely English-language articles.
- **Token volume:** ~112 long-form scholarly articles yielding ~0.8–1.2M tokens of dense, high-register Polish analytical prose.

### Author, affiliation & PII handling

- **Scholarly attribution:** Lead and contributing author names and DOIs are preserved in document metadata (`author`, `doi`, `source_url`) as official academic attribution.
- **PII scrubbing:** Author email addresses, ORCID identifiers (`orcid.org/0000-...`), phone numbers, and national identifiers (PESEL, NIP, REGON) appearing in contact blocks or bios are systematically scrubbed and replaced with stable `[PII]` tokens, preserving sentence syntax.

### Inclusion rationale & quality/bias caveats

- **Inclusion rationale:** High-quality, editorially reviewed policy research and academic discourse with unambiguous CC BY 3.0 PL licensing, permanent DOI provenance, and zero OCR noise. Fills a major gap in formal analytical Polish.
- **Quality & bias caveats:**
  - Institutional focus: Content is oriented toward state policy, parliamentary advisory topics, economics, and legal regulation.
  - Complex formatting: Rare multi-column tables, statistical charts, or boxed exhibits in PDFs may yield minor layout artifacts or omitted figure captions.

### Cross-source exact dedup — verified 2026-08-31

- All 112 released `studia_bas` texts are internally unique.
- Full byte-for-byte text comparison against 4,366,651 documents in all 20
  other local Parquets found **0 exact-text matches**.
- No `studia_bas` record required removal. A future rebuild of
  `biblioteka_nauki` still needs cross-source dedup because its current upstream
  catalogue exposes 105 of these articles even though the released local shard
  does not contain them.

---

## open_icm_pl — OPEN ICM open-access publications (pilot)

**Investigated** — 2026-09-10 (review round 2026-09-28).

**Verdict** — `added` (proposed with `release: None`). 85 publications / 2.36M
`cl100k_base` tokens — a deterministic 100-record pilot of the OPEN ICM
repository (ICM UW). Small on purpose: yields must not be extrapolated to the
6,183 eligible items.

**Enumeration** — official DSpace API (`/server/api/core/items`), fail-closed
license/language gate; repository-generated TEXT bitstreams (MD5-verified)
fetched after the origin's interactive CAPTCHA. 100 acquired -> 95 after QA ->
85 after removing 10 exact title/DOI matches against `biblioteka_nauki`.

**License evidence** — verified AT THE ORIGIN: per-item `dc.rights` +
`dc.rights.uri` in the portal's item metadata (example: "Uznanie autorstwa 3.0
Polska" -> creativecommons.org/licenses/by/3.0/pl/ on
<https://open.icm.edu.pl/items/0078ca93-45a5-4bef-912d-297c069ee6ab>, also
machine-readable at `/server/api/core/items/0078ca93-...`); 77 of 85 items
additionally ship a per-item `license.txt` bitstream (checksum-pinned).
Allowlist: CC BY / CC BY-SA only (3.0-PL / 3.0 / 4.0); NC and "dozwolony
użytek" rejected.

**Blockers** — none to merge with `release: None`. Before release: NERGAL pass -
the residual spot check FOUND one glyph-garbled contact the patterns missed
(`JKolbusz @wsiz.rzeszow.pl` in `open_icm_pl_6ec903ca-...`); extraction noise
defeats pattern matching, so the model-based scrubber is the real gate. Also
cross-source text dedup (the 10 exclusions were metadata-level) and review of
quoted third-party passages.

**Lesson** — (1) pattern scrubbers fail on PDF-extraction glyph noise - always
spot-check noisy sources, a "0 residual" pattern count is not evidence; (2)
CAPTCHA-gated origins are reproducible only via pinned checksums, not re-fetch -
say so in the datasheet instead of pretending a before/after can be regenerated.

## `1000_novels` rebuilt as its public-domain subset

**Investigated:** 2026-10-05
**Verdict:** keep 837 of 1,000 texts as `public-domain`; drop 17 in copyright and 146 unresolved.

### Verified

- CLARIN-PL (email reply, 2026-10-05): handle `11321/312` holds metadata only;
  CLARIN-PL does not distribute the texts and holds no documentation of rights in
  them. The record's CC BY 4.0 therefore licenses nothing at the text level.
- Per-text screen (`polish-dynaword-v2/evidence/1000_novels_rights_screen.csv`):
  825 public domain by term, 12 by the retained Wolne Lektury footer, 17 in
  copyright, 146 unresolved.
- Header credit lines beat names found elsewhere in the text. Four rows changed
  status (0221, 0292, 0294, 0450). `1000_novels_0981` is Seneca's letter to
  Lucilius filed under Żuławski, headed "PRZEKŁAD" with no translator named,
  so it now fails closed.
- `src/clean_1000_novels.py` writes per-row author (with translator), title, death
  year and basis to `1000_novels.attribution.jsonl`. Titles: 790 match the CLARIN
  file name, 32 are read by position (Inferred), 15 are read by hand.

### Not established

- Edition rights (Art. 99¹–99²).
- The 12 Wolne Lektury rows rest on WL's statement (Inferred).
- No dedup yet against `wolne_lektury` / `eltec_pol`.

**Lesson:** a file name is metadata, not evidence. Two texts did not match the
work their file name gives, and 4 translator credits were wrong.

## wiktionary_examples — Polish Wiktionary usage examples (rebuild)

**Investigated** — 2026-10-05 (October 2026 rights audit, `artifacts/rights_audit_2026-10/reports/wikis_ugc.md`).

**Verdict** — `added`, rebuilt for v0.2.6. Quotations from third-party works dropped;
1,818 documents / 534,521 characters (v0.2.5: 8,639 / 2,889,761). Licence relabelled
`CC-BY-SA-3.0` → `CC-BY-SA-4.0`.

**Enumeration** — official dump `plwiktionary-20261001-pages-articles.xml.bz2` (sha1
`a097f6e2…`), parsed by `src/fetch_wiktionary_examples.py`. 116,157 Polish entries;
78,245 example lines, of which 32,664 carry `{{źródło}}`/`<ref>` and 319 name a source by
hand ("(z Internetu)", "(B. Prus: Lalka)", "(Wikipedia)", Bible references, "(…)"). The
audit's sample estimated 23% cited; the full count is 42%.

**License evidence** — verified: Wikimedia Terms of Use (contributors license their text
CC BY-SA 4.0; page sha256 `913bdefc…`) and the pl.wiktionary footer linking BY-SA 4.0,
both read 2026-10-05. Inferred: pre-June-2023 text was under 3.0, whose §4(b) allows an
adaptation under 4.0. Quoted lines are used on Wiktionary under Art. 29, which does not
pass to a corpus. Evidence in `data/wiktionary_examples/wiktionary_examples.license-evidence.json`.

**Blockers** — none to merge. Quotations pasted without a source marker cannot be
detected; NERGAL pass before release.

**Unmarked-quotation sample** — kept (decided 2026-10-05) after a web check of 100 random
released lines (seed 20261005, of 7,749): no line was found copied from another source
(95% upper bound 3.7%). One line occurs on a 2017 forum post, copied from Wiktionary,
where it has been since 2007; one is a paraphrase of a 2014 news quote, not verbatim.
Web search cannot see unindexed print. Per line in
`data/wiktionary_examples/wiktionary_examples.quotation-sample.jsonl`.

**200-character gate** — kept (decided 2026-10-05). It removes 29,265 of the 31,094
entries with a kept example; several headwords per document is not done.

**Lesson** — (1) a template-only citation filter misses about 1% of quotations, which
contributors attribute by hand in parentheses; survey the uncited lines before trusting
it. (2) An end-of-section regex for `{{przykłady}}` must be anchored to line start, or an
inline `{{tłum}}` cuts the section and hides cited lines.

## nkjp1m — NKJP 1-million-word subcorpus (licence confirmed)

**Investigated** — 2026-10-05 (October 2026 rights audit). Held for v0.2.6 until IPI PAN
replied to the maintainer's question.

**Verdict** — `added`, relabelled `CC-BY` → `CC-BY-4.0`. Text unchanged:
18,381 documents; only the `license` and `author` columns differ from v0.2.5.

**License evidence** — verified: an email from IPI PAN for the NKJP Consortium (2026-10-05
22:12, Europe/Warsaw):
- the CC BY licence covers the texts as well as the annotations;
- agreements were signed with the publishers or authors of all texts, spoken ones
  included;
- the texts may be distributed without annotations.

The version is CC BY 4.0: stated on nkjp.pl (index.php?page=14, read 2026-10-05) and
confirmed in the consortium's second reply (2026-10-05). The template in the first email
named CC BY 3.0 Polska; another company had proposed it to IPI PAN for its own product. Evidence in
`data/nkjp1m/nkjp1m.license-evidence.json`.

**Attribution** — `data/nkjp1m/ATTRIBUTION.md` adapts the formula IPI PAN forwarded, which
was written for a model trained on the corpus, to redistribution of the text, and lists the
changes made. The consortium agreed to it on 2026-10-05; its licence line was changed to
CC BY 4.0 as the same reply suggested.

**Blockers** — none to merge. NERGAL pass before release.

**Lesson** — when the download page names a licence without a version ("CC-BY"),
check the project's other official pages before asking: nkjp.pl stated 4.0. A template
the licensor forwards may have been written by a third party; check its licence version.

## openstax_pl — withdrawn (2026-10-05)

**Verdict** — `removed`. Never released (`release: None`, not in the v0.2.5 tag).
1,422 documents / 4,632,374 tokens dropped. The data, the registry entry, the fetcher, its
test and requirements, and `artifacts/openstax_pl_ontology_manifest.json` were removed. A
legal-note stub (the HPLT pattern) was not kept, because nothing was released. Git history
holds the code if OpenStax Poland ever gives written permission.

**Reason** — on 2026-10-05, the "Cytowanie i udostępnianie" block on all 8 live
`/pages/przedmowa` pages allowed reuse only for non-commercial purposes. It also barred
LLM training without OpenStax Poland's prior written consent. The OpenStax CMS lists all
8 Polish books as CC BY-NC-SA 4.0. The sources contradict each other: the same block
labels itself "(CC BY)", and openstax.pl's FAQ and catalogue say CC BY 4.0. Whether the
book-level wording was live at our 2026-09-05 fetch is not established; the last archived
captures predate it. The audit read this as unresolved and restrictive, which is not
enough to ship under a CC BY label. References:
`rights_audit_2026-10/reports/academic.md` L3-9; `reports/citations/openstax_cv.md`
A1-A14 and Corrections 1-5.

**Cause** — the fetcher read only the foreword body and hard-coded `CC-BY-4.0`. It never
read the licence block, the site footer or the CMS licence field.

**Lesson** — the licence on the badge or in the body is not the whole grant. Read every
rights statement the publisher shows, plus its machine-readable licence field, and stop
on any NC, ND or AI-training term (`.agents/skills/source-rights-review`, check 4).

## fandom_pl — withdrawn (2026-10-05)

**Verdict** — `removed`, whole source (all 12 wikis). Never released (`release: None`,
not in the v0.2.5 tag, not in the README totals). 50,653 documents / 45,790,009 tokens
dropped. The data, the registry entry, `src/build_fandom_pl.py`, its test and
requirements, and `artifacts/fandom_pl_ontology_manifest.json` were removed. No
legal-note stub was kept, because nothing was released. This decision supersedes the
earlier audit action "drop 6 wikis, sample 4".

**Reason** — (1) The Fandom Terms of Use (revision 2025-12-19, read from a Wayback
capture of 2026-09-05) bar three things without express written permission: scraping
and automated access, bypassing robot-exclusion headers, and using content to train an
ML/AI system. The AI clause is not in the 2023 revision and has no carve-out. Whether it
overrides the CC BY-SA grant is a legal question we have not resolved. (2) The pages were
fetched live from each wiki's MediaWiki API on 2026-10-02, after that revision. The
builder has no robots.txt handling. `PiotrSty/fandom-pl` is the preparer's own
repository, not an independent snapshot. (3) In 40-document samples, the wikis carry
third-party text under the CC BY-SA label. Affected documents: dragonage 22/40 (game
codex, letters, poems; EA/BioWare), elderscrolls 17/40 (full English in-game books;
Bethesda), wiedzmin 11/40 (CD PROJEKT RED bars text and data mining). naruto has lyrics
pages. pokemon includes Bulbapedia translations, and Bulbapedia is CC BY-NC-SA 2.5.
creepypasta has 240 labelled translations of stories whose origins are mostly
untraced. (4) All 12 wikis report an unversioned `CC-BY-SA` in rightsinfo. References:
`rights_audit_2026-10/reports/wikis_ugc.md` L12-31; `reports/citations/fandom.md`.

**Lesson** — a site's CC licence for user contributions does not settle how its content
may be fetched. Read the host's terms for scraping and AI-training clauses, and the
revision date, before a live API crawl. Fan wikis also quote the works they cover;
sample them for game text, lyrics and translations before trusting the site licence.

## common_voice_pl — withdrawn (2026-10-05)

**Verdict** — `removed`, whole source. Never released (`release: None`, not in the v0.2.5
tag). 45,043 sentences / 994,922 tokens dropped. The data, the registry entry,
`src/fetch_common_voice_pl.py`, its contract test,
`artifacts/common_voice_pl_ontology_manifest.json` and
`artifacts/common_voice_pl_language_audit.json` were removed. No legal-note stub was
kept, because nothing was released. The audit's alternative, dropping only the Europarl
sentences, was not chosen. The open question of sentence-level units is moot.

**Reason** — CC0 is Common Voice's platform-level label; the sentences were never shown
to be CC0. Matching the shard against Common Voice's Polish sentence files gave these
counts: 24,709 (54.9%) are in `selected-europarl-v7-pl.txt`, 18,569 (41.2%) are in
`sentence-collector.txt`, 37 are in both, and 1,802 (4.0%) are in neither. The counts
reproduce only with NFKC + casefold + removal of all non-word characters
(`openstax_cv.md` B.3). The recipe first stated in the audit gives different numbers.
- Europarl: the European Parliament legal notice allows reuse with source
  acknowledgement, and asks for the URL of the complete item when only part is
  reproduced. The shard carries neither, and an attribution condition cannot sit under
  CC0. Mozilla's 2020 legal review of Europarl was never shown as completed. The Common
  Voice README's CC0 sentence does not cover the Europarl file.
- Sentence Collector: volunteers attest these lines are public domain, but their sources
  are not recorded. `sentence_domain` is "unspecified" for 45,035 of 45,043 rows.

References: `rights_audit_2026-10/reports/citations.md` L35;
`reports/citations/openstax_cv.md` section B, Corrections 6-13.

**Lesson** — a dataset-level CC0 is only as good as the per-item provenance behind it.
Match the items against the platform's own source files before trusting the label. A
public-sector text with an attribution condition cannot be relabelled CC0.

## wikiquote — withdrawn (2026-10-05)

**Verdict** — `removed`, whole source, from main and from v0.2.6. Released in v0.2.0 and
in every tag through v0.2.5, which stay as published. 30,363 documents / 31,896,591
tokens dropped. As with HPLT, the registry keeps its `release: "0.2.0"` entry, now with
a `legal_note`; `make_docs.py` skips it (no stats file). `plwikiquote` was removed from
`src/fetch_wikimedia.py`, so a default run no longer fetches it. The audit's
alternative, screening per page for authors and translators who died in 1955 or
earlier, was not chosen; its 10-20% yield estimate was not reproduced (Q14).

**Reason** — most of the text quotes other people's works. CC BY-SA covers the
contributors' selection and wording, not the quotes. The quotes rest on the Art. 29
quotation right (t.j. Dz.U. 2025 poz. 24), which is purpose-bound and is not a licence
to downstream users; its reach to bulk reuse is inferred, not verified.
- Random 200-page sample (seed 1), audit's own labels: 158 in copyright (79%, Wilson
  72.8-84.1%), 20 public domain, 13 mixed, 7 translation-only, 2 unknown. The citation
  pass checked only the arithmetic, not the labelling (Q13).
- Example: `wikiquote_2429270` is the page on Alice Sebold (b. 1963), quoted in modern
  translations (Q10-Q12).
- The 1,000-character quote cap is a length guideline with exceptions up to 1,800
  characters, not a copyright rule (Q1-Q3). No pl.wikiquote copyright-policy page was
  found (Q8).

References: `rights_audit_2026-10/reports/wikis_ugc.md` L1-4; `reports/citations.md`
L11; `reports/citations/wikimedia.md` section 3 (Q1-Q14); sample in
`rights_audit_2026-10/wikis/wq_sample.json`.

**Lesson** — a project licence covers what the project's contributors wrote. On a
quotation site that is the framing, not the quoted text.

## wikisource — withdrawn (2026-10-05)

**Verdict** — `removed`, whole source, from main and from v0.2.6; a rebuild for a later
release is tracked separately (polish-dynaword-v2 #68). Released in v0.2.0 and in every
tag through v0.2.5, which stay as published. 632,005 documents / 801,947,427 tokens
dropped. As with HPLT, the registry keeps its `release: "0.2.0"` entry with a
`legal_note`. Its `traceable` text no longer claims every text is public domain, and
its `upstream` is now pl.wikisource.org rather than the dumps it never came from.
`count_vocab.py` and `sample_quality.py` no longer list it. There was no fetch script
to remove: the data was a SpeakLeash pull.

**Reason** — all rows carry one `CC-BY-SA-3.0` label, and the card said the underlying
texts are public-domain works. Wikisource's own guidance says "mostly" (W1).
- Among the 12 audited rows, several are in-copyright post-war imprints that their
  rights holders released through consent programmes or OTRS, each page under its own
  licence template: PIW 1971 and 1972, Czytelnik 1962 and 1966, WL 1974, a 1995 sailing
  manual, 1990-91 regional press (W11-W21). The templates differ (CC BY 3.0 PL,
  CC BY-SA 3.0 PL, CC BY-SA 3.0, GFDL, CC BY 4.0), so no single relabel fixes them (W17).
- One row carries a translator's footnote for a translation still protected until 2045;
  the row-to-page mapping is inferred (W8-W10).
- The rows keep neither the consent, the licence version nor an author (W22). The
  SpeakLeash selection is undocumented and rows have no title or page id, so the rows
  cannot be screened afterwards. The audit's crude filter (372 rows, 1.7%) was not
  reproduced; the in-copyright share is unknown (W23).

References: `rights_audit_2026-10/reports/citations.md` L10;
`reports/citations/wikimedia.md` section 2 (W1-W23); `REVIEW.html` (wikisource block).

**Lesson** — a project-wide label on an aggregator pull hides per-page licences. Pull
from the official dump with page ids and licence templates, so each page can be
screened and labelled.

## biblioteka_nauki — public-domain-label records dropped (2026-10-06)

**Verdict** — `added`, kept for v0.2.6 minus 3 records (polish-dynaword-v2#52). 42,071 →
42,068 documents, 673,488,831 → 673,112,676 tokens (−376,155), 1,748,333,935 →
1,747,387,478 characters. `src/clean_biblioteka_nauki.py` filters the v0.2.5 Parquet
(pinned by SHA-256): it keeps `cc-by`, `cc-by-sa` and `cc0`, drops `public-domain`, and
fails on any other value. `recover_attribution.py` no longer counts `public-domain` as clean.

**Reason** — the 3 rows (`books_2142716`, `books_2147851`, `articles_2097363`) carry the
platform's "Domena publiczna" label. Verified 2026-10-06 against
`bibliotekanauki.pl/api/{type}/{id}`: licence id 22, `creativeCommons: false`, URL
`creativecommons.org/publicdomain/mark/1.0/`. Creative Commons describes the Mark as "a tag
or a label" for works already free of copyright; it grants nothing. All three are original
texts (`reprintPublication` and `translationPublication` false) first published 2019, 2021
and 2022 by named authors (Jerzy Zając, Witold Rakowski, Beata Michalec), so life+70 has
not run out. The 2019 book's own colophon reads "Copyright by Jerzy Zając 2019 / Copyright
by Muzeum Niepodległości w Warszawie 2019". That the authors are alive is inferred.

**Not done here** — the licence version is still collapsed: rows say `cc-by`/`cc-by-sa`,
the card and `stats.json` say 4.0. A live re-check (seed 20261006, 24 rows) found 4 of 12
`cc-by` rows under 3.0 PL or 3.0 Unported; all 12 `cc-by-sa` rows were 4.0. Fixing it needs
a per-record API re-fetch of name and version (tracked under polish-dynaword-v2#57). The
card still names SpeakLeash while `src/sources.py` says direct rebuild; `licenses.jsonl`,
`index.jsonl` and `build_parquet.py` are not in the repo. 128 kept rows (88 `cc-by`, 40
`cc-by-sa`; 1.56% of tokens) contain "Wszelkie prawa zastrzeżone" or "All rights
reserved" in their text (publication years 2010-2021); regex count only, not reviewed.

**Lesson** — a platform's "public domain" label can be the PD Mark. Check the URL behind
the label, and check the date: a 2019 work cannot be public domain by term in Poland.

## govpl — PAP reprints dropped (2026-10-06)

**Verdict** — `kept`, minus 188 PAP reprints (slayerlabs/polish-dynaword-v2#51).
88,190 → 88,002 documents, 80,666,129 → 80,445,758 tokens (−220,371). `release` stays
`0.2.2`. `src/clean_govpl.py` filters the pinned v0.2.5 Parquet, so ids are unchanged,
and writes `govpl.attribution.jsonl` with each row's gov.pl URL, subsite, title and date.
`fetch_govpl.normalize()` applies the same rule (`pap_rule`) to future crawls.

**Reason** — gov.pl's copyright page grants CC BY-SA 4.0 for its text but carves out
works whose copyright belongs to third parties; their use needs the owners' consent.
Articles that credit PAP as their source are such works.
- The issue's rule (`Źródło:` line naming PAP) matches 154 rows. A hand-check of 20
  (seed 51) found 20 text source lines and no photo credits; 14 credit PAP jointly with
  the ministry, so the text may be partly the ministry's. PAP's archive is not public, so
  verbatim PAP authorship was not confirmed.
- A scan of the other rows added 34 verified text attributions: misspelt or colon-less
  source lines (9), `Na podst. … PAP` (22, all uw-podlaski), a lone `(PAP)` sign-off line
  (2) and one PAP interview transcript (prokuratura-krajowa). Photo credits (`fot. PAP`)
  and rows that quote or mention PAP stay.
- Left in, flagged: a mid-text `(PAP)` after one paragraph (govpl_30227, govpl_58869), a
  footnote crediting PAP and the health ministry (govpl_81037), a KRRiT report quoting two
  PAP questions (govpl_86914).
- Not reviewed: 710 kept rows carry a `Źródło:` line naming some other source (mostly
  government bodies; some name WHO, universities, journals).

References: `rights_audit_2026-10/reports/govweb.md` (govpl block);
https://www.gov.pl/web/gov/prawa-autorskie (read 2026-10-06).

**Lesson** — an open-licence footer covers the publisher's own text. On a news-style
site, check source lines for agency copy before taking the footer at face value.

## samorzad_gov_pl — third-party slices dropped (2026-10-06)

**Verdict** — kept, relabelled, 144 rows dropped: 72,472 → 72,328 documents,
40,011,099 → 39,890,244 tokens (baseline: main after #46, not the v0.2.5 tag's
72,771). The basis is now called an operator-declared platform licence. A question
to KPRM is pending (polish-dynaword-v2 #49); its answer may still change the
source's status. The drops are in `src/clean_samorzad_gov_pl.py`
(`third_party_drop`, applied by `--refilter-release` to the pinned release):
- 69 "Piątek z poezją" posts (all Powiat Sokołowski): poems credited to their
  authors. All 69 hand-checked.
- 9 IMGW-PIB notices carrying "Opracowanie niniejsze i jego format…", which forbids
  changes and requires naming IMGW-PIB. All 9 hand-checked.
- 66 articles whose `Źródło:` line credits a non-governmental source (Wikipedia,
  media, associations including volunteer fire brigades, foundations, clubs, rural
  housewives' circles, companies, named books or brochures), unless a photo or film
  marker precedes it on the line. All 66 hand-checked as third-party text. The rule
  was revised on the same matches (first pass 61 of 67 correct), so precision is
  in-sample and recall is unknown; one known miss is a Fundacja Santander credit.
  Part of the regex lists work titles found in that review.
- Not dropped: contest entries. The issue's 112 is not reproducible; the closest
  regex (awarded/winning/distinguished works) gives 111 documents, and a 40-document
  sample (seed 49) contains no entry text: they are results announcements. Only 3
  documents publishing entries were found; most entries are images, not text.

**Reason** — gov.pl's copyright page says the service also carries third-party works
whose use requires the owners' consent; the footer licence cannot cover them.
Credits matter here: 810 documents have an `Źródło:` line, but in 405 every credit is
only a de-identified `<URL>`, so their sources cannot be classified; 231 documents
mention IMGW without the verbatim notice and were not reviewed.

References: polish-dynaword-v2 #49;
`rights_audit_2026-10/reports/govweb.md` L8-15; `reports/citations/samorzad_nonsa.md`
S1, S8, C4; <https://www.gov.pl/web/gov/prawa-autorskie> (read 2026-10-05).

**Lesson** — de-identification that replaces URLs also removes the provenance needed
to screen reposts. Record the credited domain before masking URLs.

## global_voices — partner republications dropped (2026-10-06)

**Verdict** — `admitted` with 114 posts withheld (polish-dynaword-v2 #50). The source
goes from 2,040 docs / 10,073,628 chars / 3,715,020 tokens to 1,926 / 9,358,071 /
3,453,918. The posts come from 51 outlets. The new `src/clean_global_voices.py` filters
the v0.2.5 Parquet and sidecar, pinned by SHA-256. `src/test_global_voices_contract.py`
checks that no shipped row matches the rule.

**Reason** — Global Voices licenses "content created by Global Voices" under CC BY 3.0.
A post that republishes a partner's article under a content-sharing agreement or with
permission stays the partner's work.
- **Where the rule looks.** It reads only the first 700 and last 350 characters, where
  the editorial note sits. A rule fires only on a sentence about the post itself that
  does not name Global Voices, so news reporting ("raport opublikowany przez UNESCO")
  and GV's own projects do not count. Sentences whose subject is an image are skipped
  too.
- **Rule counts** (a post can match several):
  - originally: 69
  - agreement: 56
  - published_by: 51
  - partnership: 11
  - permission: 5
  - republished: 5
  - written_for: 3
  - non_commercial: 3
  - listed by hand: 3 (21466, 27005, 32352)
- **Hand check.**
  - Every one of the 114 drops was read, and the English original was consulted where
    the note was unclear.
  - 25 rule matches are kept by hand with a stated reason:
    - 12 cross-posts from the author's own blog;
    - 7 posts GV wrote for the ISN blog;
    - 3210, a GV post that another outlet republished;
    - 2 notes dating GV's own English original;
    - 3 news mentions.
  - All 43 posts where the gating removed a raw match were checked. They are image
    credits, news and GV projects.
  - 15 partner-organisation accounts with no republication note (e.g. 350.org 22567,
    Metamorphosis 33071) were checked against the English page and kept.
- **Outlets.**
  - PRI/The World 21
  - The Stand News 8
  - Hong Kong Free Press 7
  - Agência Pública 6
  - Metamorphosis/Meta.mk 5
  - 350.org 5
  - Eurasianet 3
  - Freemuse 3
  - The Irrawaddy 3
  - 2 each: Balkan Diskurs, Ensia, Kurtural, Loa, MigraMundo, NACLA, Nepali
    Times/Himalkhabar, Premium Times, Syria Untold, Teplitsa, Transitions
  - 1 each: 88 Bar, APC.org, Afroféminas, Azerbaijan Internet Watch, Baynana, Bumaga,
    CHRI, Cari-Bois, Chaikhana, Common Dreams, Connectas, Doha News, Fundación MEPI,
    GCHR, Hive-Mind/EJO, Iran Voices, New America Media, OC Media, Periodismo de
    Barrio, Promundo, Russia Post, Sin Embargo, Syria Deeply, TJournal, Tea After
    Twelve, The Isaan Record, The Moscow Times, VideoVolunteers, WITNESS,
    inmediahk.net, openDemocracy.
- **Delta against the audit's 64.** The audit used outlet names plus an agreement
  regex, and it recorded neither the list nor the script. Its "17 PRI" is an
  undercount: all 21 posts credited to PRI/PRX's The World carry a republication note.
  Most of the remaining difference is outlets the audit's name list did not cover,
  such as The Stand News (8) and Agência Pública (6). The Hub PR has per-document
  evidence.
- **Borderline; dropped and flagged for review.**
  - Freemuse ×3: the note allows republication "by non-commercial media" only.
  - Bumaga 30003: a shorter version of the author's article for Bumaga.
  - Hive-Mind/EJO 34820.
  - openDemocracy 29544: "under a Creative Commons licence", version not stated.
  - Transitions 34277: "powstał dzięki umowie partnerskiej", possibly a co-production.
  - 32352: built on two Meta.mk texts.

**Lesson** — the licence covers what the publisher created. A per-post editorial note
is the only signal of a partner's work, so the filter has to read the note and not only
the byline.

## `1000_novels` rescreened: unknown authors fail closed, translator death years cited

**Investigated:** 2026-10-06
**Verdict:** keep 888 of 1,000 texts as `public-domain` (was 837); drop 18 in copyright
and 94 unresolved. Tracks slayerlabs/polish-dynaword-v2#44.

### Verified

- Screen defect: a known public-domain translator cleared an author whose death year
  was unknown. `1000_novels_0733` (author "Roux", unidentified) shipped in the
  first rebuild this way. An unknown author now fails closed unless the work is listed
  as anonymous.
- Translator death years now come from Biblioteka Narodowa / VIAF authority records,
  cited in `polish-dynaword-v2/scripts/novels_rights_screen.py`. Józef Pracki d. 1882,
  Edmund Cięglewicz d. 1928, Stanisław Sierosławski d. 1941, Bolesław Londyński d. 1928
  (22 texts), Ludwik Jenike d. 1903, Adam Stögbauer d. 1916, Kazimierz Bronikowski
  d. 1909, Maria Cumft-Pieńkowska d. 1920, Izydor Berman d. 1942, Konrad Drzewiecki
  d. 1922, Stefan Frycz d. 1960.
- "tłum. A. W." on 29 Chekhov texts is Antoni Lange (d. 1929): BN catalogues their
  source edition (*Śmierć urzędnika*, Bibljoteka Groszowa 1926) with "nazwa właśc. tł.:
  Antoni Lange".
- `1000_novels_0981` (Seneca's letter, "PRZEKŁAD", no name) is Żuławski's own
  translation in his 1914 collection *Bajka o człowieku szczęśliwym*. It is now
  credited to Seneca with Żuławski as translator.
- "Roux" is Saint-Pol-Roux (d. 1940), so `0733` stays in on a cited death year.
- `1000_novels_0462` (*Wola mocy*): BN credits Stefan Frycz (d. 1960) as
  co-translator. It is dropped although its Wolne Lektury footer says public domain. The
  screen's footer cross-check allowlists it by name.
- Net: 53 unresolved and 10 Wolne-Lektury-footer rows move to public domain by term;
  1 Wolne-Lektury-footer row moves to in copyright.

### Not established

- "A. L." (`1000_novels_0734`, Schopenhauer 1901): BN records the cryptonym only. The
  row still rests on its Wolne Lektury footer (Inferred).
- Edition rights (Art. 99¹–99²); dedup against `wolne_lektury` / `eltec_pol` (#45).

**Lesson:** a Wolne Lektury public-domain footer is a claim, not evidence. One of the
12 it covered had a translator who died in 1960.

**Update (2026-10-06, maintainer decision):** the screen no longer keeps a row on a
Wolne Lektury public-domain footer. "A. L." (`1000_novels_0734`) now fails closed as
unresolved, so the release keeps 887 texts and drops 18 in copyright and 95 unresolved.
`src/clean_1000_novels.py` pins the screen CSV by SHA-256
(`7fefc75f8fcbd07e5ec98842276765ffef9520b6acb4e24b4751c67cc3d2081c`, also in
`1000_novels.license-evidence.json`).

## wolne_lektury — eltec_pol duplicates dropped (2026-10-06)

**Verdict** — kept, 63 documents dropped: 6,141 → 6,078 documents, 103,009,797 →
90,654,820 tokens (input: the v0.2.5 Parquet, unchanged since v0.2.1). Each dropped
document is a novel, or one tome of a novel, that `eltec_pol` also carries. 50 ELTeC
novels are concerned, and `eltec_pol` stays whole (maintainer decision A,
polish-dynaword-v2 #72). The filter is `src/clean_wolne_lektury.py` and the pair list is
`data/wolne_lektury/wolne_lektury.eltec_pairs.tsv`.
- Score: the share of a WL document's 5-word shingles found in the ELTeC novel that
  shares the most with it. Tokenisation follows `near_dedup`, and a trailing WL footer is
  cut first (no WL row has one). Containment is used, not Jaccard, because WL splits
  novels into tomes: Chłopi has 4, Faraon, Nad Niemnem and Stara baśń have 3.
- Threshold 0.9, in an empty stretch: no document scores between 0.7588 and 0.9636. All
  63 documents above it pair with an ELTeC novel by the same author and title (Żuławski's
  two WL documents are headed with the trilogy's title).
- Hand-checked: all 3 documents between 0.3 and 0.9, the 3 between 0.9 and 0.99, and 10
  sampled at ≥ 0.99 (seed 72). In 12 of the 13 drops checked, the unmatched shingles
  form short scattered runs of at most 27 words (title head, spelling variants). The
  exception is Dzieje grzechu (0.9879): its WL copy ends with a "Dodatek" of draft
  chapters, about 1,800 words that ELTeC lacks. It is dropped with the novel.
- Kept and listed as `keep_partial`: both WL tomes of Mniszek's Trędowata (0.7588 and
  0.6473 against POL0051). The ELTeC text is shorter: about 121,000 words against WL's
  162,000. 95% of the ELTeC shingles are in WL's tomes, and the WL text missing from
  ELTeC comes in runs of up to 5,418 words. The shared part stays duplicated across the
  two sources. Also kept: Kochanowski's Pieśń IX from Pieśni, księgi wtóre (0.4267),
  which Łoziński's Oko proroka quotes.
- Below 0.3 (not listed, not dropped): 6 documents at 0.10–0.22, all poems of at most
  2,321 tokens by Mickiewicz, Kochanowski and Miciński; their overlap with the paired
  novels was not inspected. Nothing scores between 0.2203 and 0.4267.

**Reason** — 51 of the 100 ELTeC novels duplicate WL texts. ELTeC's CC BY 4.0 licence
is the clean one (`rights_audit_2026-10/reports/literary.md`: eltec_pol CLEAN). WL
licence labels are not touched here; #57 relabels them per row.

References: polish-dynaword-v2 #72 (decision), #45 (method);
`rights_audit_2026-10/literary/eltec_meta.tsv` (ELTeC file names and titles).

**Lesson** — Jaccard near-dedup misses a novel that one source splits into tomes.
Measure containment from the split side.

## 1000_novels — wolne_lektury and eltec_pol duplicates dropped (2026-10-06)

**Verdict** — kept, 666 texts dropped: 887 → 221 documents, 130,201,301 → 26,625,861
characters, 51,981,548 → 10,759,728 tokens (input: Hub PR 65; references: main at
563eea8, after Hub PR 67). Each dropped text is already in `wolne_lektury` (629) or
`eltec_pol` (37); none is in both, since Hub PR 67 dropped the WL copies of ELTeC novels
(before it, 35 of the 37 were in both). Maintainer decision on polish-dynaword-v2 #45:
drop the 1000_novels copy and keep both other sources whole. The filter is the dedup
step of `src/clean_1000_novels.py`, run after the rights screen. Covers and matched ids
for all 887 screened texts are in `data/1000_novels/1000_novels.dedup.tsv`.
- Cover: the share of a text's 5-word shingles found in any document of one source
  (`src/shingle_containment.py`, shared with `src/clean_wolne_lektury.py`; near_dedup
  tokenisation; a trailing Wolne Lektury footer cut first). Scored per source, over the
  union of its documents: 7 texts are duplicates only across several WL documents. For
  example, *Popioły* (`1000_novels_0953`) is in 3 WL tomes, none above 0.33, together 0.9602.
- Threshold 0.5, in empty stretches: no text covers between 0.0727 and 0.5178 against
  `wolne_lektury`, or between 0.0462 and 0.7771 against `eltec_pol`. Both reference
  Parquets are pinned by SHA-256 (main's: wolne_lektury after Hub PR 67, eltec_pol
  unchanged since v0.2.1).
- Hand-checked every text between 0.3 and 0.7, plus `0703`. All are dropped.
  - `0255` Grabiński, *Spojrzenie* (0.6721): the same story. Soft-hyphen word breaks
    ("opuści ła") lower the cover evenly; no unmatched run exceeds 20 shingles.
  - `0388` Laclos, *Niebezpieczne związki* (0.6975 over two WL tomes, 0.41 and 0.29): the
    whole 1912 Boy translation in old spelling, which lowers the cover throughout (0.51–0.88
    per tenth of the text). The longest unmatched run is the collected footnotes at the
    end (420 shingles).
  - `0509` Orzeszkowa, *Bóg wie kto* (0.6291): a 1909 serial instalment, 2,342 of WL's
    6,677 words, in old spelling; no unmatched run exceeds 28 shingles.
  - `0891` Verne, *Przygody trzech Rossyan…* (0.5383): the same Anczyc translation. Old
    spelling ("é") lowers the cover evenly; no unmatched run exceeds 49 shingles.
  - `0703` Reymont, *Chłopi* (ELTeC 0.8259; WL 0.8272 before Hub PR 67 dropped its 4
    tomes, 0.0030 after): an abridged edition of about 43,000 words in old spelling; no
    unmatched run exceeds 25 shingles.
  - `0493` Orkan, *Listy ze wsi* (0.5178 against WL's *Płanety*): sections I and II match
    at 0.76–0.90 per tenth of the text. The last section (IV, about 1,100 words) is not in
    WL and is lost with the drop.
  - 10 drops sampled with seed 45 (all WL): 98.3–100% of each text's shingle positions
    match; the longest unmatched run is 85 shingles (the footnotes of `0508`).
- Kept texts: none carries a Wolne Lektury footer. 4 share author and title with a WL
  document but are other texts: Balzac's *Kobieta trzydziestoletnia* (`0065`, Marrene's
  translation; WL has Boy's; 0.0388), Grabiński's *Na wzgórzu róż* (`0239`, 0.0043),
  Kraszewski's *Dziad i baba* (`0359`, a prose tale; WL has the poem; 0.0079) and
  Przybyszewski's *Nad morzem* (`0690`; WL has a 354-word piece; 0.0303).
- Hub PR 67 (#72) merged first. Re-pinned to its wolne_lektury Parquet (SHA-256
  `f308d492…`, was `5833c25d…`) and rebuilt: the same 666 drops and the same Parquet. The
  split went from WL 664, ELTeC 37, both 35 to WL 629, ELTeC 37, both 0; 216 WL covers
  change. The 5 texts hand-checked between 0.3 and 0.7 and the 10 sampled drops stay WL
  drops.

**Reason** — 666 of the 887 public-domain texts, 41.2M tokens, are copies of documents
that `wolne_lektury` or `eltec_pol` already release.

References: polish-dynaword-v2 #45 (measurement and decision), #72 (wolne_lektury side).

**Lesson** — score against the union of a source's documents. Wolne Lektury splits
novels into tomes, and no single tome need hold half of the text.

## wikipedia — Polish Wikipedia (rebuild from the official dump)

**Investigated** — 2026-10-05 (October 2026 rights audit, `artifacts/rights_audit_2026-10/reports/wikis_ugc.md`
L44-46, `reports/citations.md` L17, `reports/citations/wikimedia.md` section 5;
slayerlabs/polish-dynaword-v2#69).

**Verdict** — `added`, rebuilt for v0.2.6 from the official Wikimedia dump; the SpeakLeash
`plwiki` intermediary is removed. 1,468,656 documents / 3,140,841,847 characters /
1,226,913,508 tokens (v0.2.5: 1,171,897 / 1,844,707,761 / 707,194,207). Labels per record by
the page's last revision: `CC-BY-SA-3.0` 190,837 documents (6.6% of tokens), `CC-BY-SA-4.0`
1,277,819 (93.4%); v0.2.5 labelled every row 3.0. Row ids are now `wikipedia_<page id>`
("Bitwa pod Grunwaldem": `wikipedia_25295` in v0.2.5, `wikipedia_716` now).

**Enumeration** — `plwiki-20260901-pages-articles`, 7 shards, each sha1 matched to the
dump's `dumpstatus.json` and `sha1sums.txt`, parsed by `src/fetch_wikipedia.py`: 2,877,278
pages → 2,304,425 in namespace 0 → 598,071 redirects and 35 pages with `{{NPA}}`
(`{{Copyvio}}`, `{{Plagiat}}`) skipped → 1,706,319 converted → 382 exact duplicates →
231,717 under 200 characters and 5,564 non-Polish dropped by `build_dynaword.py` →
1,468,656. One sidecar line per row (`wikipedia.attribution.jsonl`: page id, revid, title,
URL, last-revision time, licence).

**License evidence** — verified, read 2026-10-05: Terms of Use oldid 554823 (CC BY-SA 4.0
and GFDL, in effect June 7, 2023; sha256 `c96ef93b…`), oldid 98524 (CC BY-SA 3.0, June 16,
2014), the 3.0 and 4.0 legal code, pl.wikipedia `rightsinfo` (BY-SA 4.0). Inferred: a page
edited after the switch may carry its earlier 3.0 text under 4.0 (3.0 §4(b)(ii)). The
cutoff is 2023-06-07 00:00 UTC, not the audit's "06-08 if time of day unknown": 06-08 would
put that day's 4.0 edits under 3.0, which 4.0 §3(b)(1) does not allow; midnight may put
3.0 text under 4.0 for some of the 184 pages last edited that day. The footer switch date
was not checked. Evidence in `data/wikipedia/wikipedia.license-evidence.json`.

**Copyright pastes** — the 31 notice pages unchanged since the dump were re-read: all carry
the notice; 4 edited since were not re-read. Web sample of 40 released pages per label
(seed 20261006, Earwig's Copyvio Detector, every hit from confidence 0.10 read by hand):
1 of 40 per label copied or closely paraphrased from a copyrighted source
(`wikipedia_833396` "Obnażyć prawdę", Filmweb's official description; `wikipedia_1126780`
"Czarnka", the Drawieński Park Narodowy lake text archived 2012), 6 later copies of
Wikipedia, 2 shared titles or stock phrases, 70 below the threshold. 95% Wilson upper
bound 12.9% per label. Both pages are kept in this release; dropping them is an open
decision. Per page in `data/wikipedia/wikipedia.copyvio-sample.jsonl`.

**Dedup checks** — reported, not fixed. Exact-text overlap (SHA-256 of raw text) with NKJP1M
(main and v0.2.5) and HPLT v0.2.5 is 0 documents (v0.2.5: NKJP1M–Wikipedia 1, HPLT–Wikipedia
6). Near duplicates not measured. The `overlap_audit.json` files of `plwiki_talk` and others
compare literal ids against the old `data/wikipedia` (revision `fe44303`, 1,171,897 rows) and
are stale after the id change.

**Blockers** — none to merge. Open: the two pasted pages; the time of day of the 4.0 switch;
GFDL-era text; `{{ek}}` pages kept; appendix headings left in 56.4% of documents (2.96% of
characters from the first such heading on) and whitespace-only lines in 11.9%; NERGAL pass
before release.

**Lesson** — (1) Earwig rated a close paraphrase "none" (0.30): hand-check below its
"possible" threshold. (2) Earwig's best hit is usually a later copy of Wikipedia; the real
source came from the article's own citation and history ("Czarnka": a 2025 forum hit, the
park's 2012 text behind it). (3) The Terms of Use give the day of the licence switch, not
the time, so a last-revision label is inexact for one day of edits.

## wikipedia — update: copied pages and `{{ek}}` dropped

Drops added to `src/fetch_wikipedia.py` (`prepare`), keyed on the wiki page id; the dump
(plwiki 20260901) is unchanged. `COPIED_PAGES`: page 833396 "Obnażyć prawdę" (Filmweb's
official description; excluded) and page 1126780 "Czarnka" (Drawieński Park Narodowy lake
text; held for review, Art. 4 not assessed). The build fails if a listed page is not reached.

`{{ek}}` (speedy deletion; aliases EK, Ekspresowe kasowanko, Usuń, Speedy, Delete from the
dump's redirects) is read with its reason; a reason naming a copyright violation (npa,
plagiat, copypaste, kopiuj-wklej, copyvio, "praw… autorsk…") drops the page. The dump has 0
live `{{ek}}` notices on articles, so 0 dropped and no reasons to count; a looser text scan
of all 7 shards found 3, all inside `<!-- -->` comments (redirect, content moved, no
reason), none about copyright. The hand-check of 10 was not possible.

Rebuilt from the same dump: 1,468,656 → 1,468,654 documents, 1,226,913,508 → 1,226,912,191
tokens (3.0: 190,836 docs / 81,022,938 tokens; 4.0: 1,277,818 / 1,145,889,253). Diff against
the previous parquet: only `wikipedia_833396` and `wikipedia_1126780` removed, no row added
or changed, same in the attribution sidecar. Row order differs between builds
(`build_dynaword.py` uses `imap_unordered`); content per id is identical. The copyright
sample keeps its population (the previous build) and records the two dropped ids.

## sejm_api — provenance relabelled (2026-10-06)

**Verdict** — kept, rows unchanged: 38,812 documents, 36,335,587 tokens, parquet
sha256 `56f7b03483488dec8f0fa145c6d26746926175d83096e9d1931e535aa81cf161` as in v0.2.5.
Registry, card and NOTICE now say "Sejm API data via contributor snapshot" instead of
"Direct official Sejm API records". The text is the Hugging Face dataset
`PiotrSty/sejm-speeches-corpus@v1.0.1` (commit `124557799608`); SlayerLab did not
fetch it from the API.

**Reason** — all 38,812 `source_url` values in the sidecar match
`https://api.sejm.gov.pl/sejm/term{N}/videos/{date}`: one URL per sitting day, so
the "speech-level provenance" line in the card was wrong. That is now a test in
`src/test_sejm_api_contract.py`.

### Not established

- Whether the API publishes per-speech URLs, and whether the snapshot text matches
  the live API (the audit asked for 20 rows to be spot-checked; not done here).
- Whether the www.sejm.gov.pl terms reach API output; the API documents no licence.
- (Fixed in a follow-up commit: the card's "Temporal novelty" paragraph named
  `parlamint_pl` as registered; it left the registry in #54.)

References: polish-dynaword-v2 #57; `rights_audit_2026-10/REVIEW.html` `#sejm_api`;
`reports/citations/sejm_sis.md`.

**Lesson** — a card line that says "speech-level" must be backed by a test on the
sidecar, not by the template that wrote it.

## `open_agh_chemistry_pl`: rights pages pinned, no archive copy exists

**Investigated:** 2026-10-06
**Verdict:** no data change; 370 documents stay CC BY-SA 4.0. The licence evidence is now in
`data/open_agh_chemistry_pl/open_agh_chemistry_pl.license-evidence.json`. Tracks the 2026-10
rights audit action "archive the rights pages".

### Verified

- Re-fetched the four EPUBs (`/rpc/preview/handbooks/{29,1394,37,1893}/epub/light`) on
  2026-10-05 23:38 UTC. EPUB and `OEBPS/main.xhtml` sha256 equal the values recorded on
  2026-09-05, 4/4.
- Each rights page says "Publikacja udostępniona jest na licencji Creative Commons Uznanie
  Autorstwa - Na tych samych warunkach 4.0." and links the 4.0 deed. No NC marker, no "3.0".

### Not established

- No Wayback snapshot of any rights-bearing URL exists (CDX lookups, read-only; no Save Page
  Now). The `/handbook/{id}` landing pages are archived, but they are a ~3 kB shell without
  licence text. A later relabel of the rights pages cannot be ruled out; the bytes are pinned
  by sha256 instead. A written confirmation from AGH would close the gap.
- Whether AGH holds every module's rights (no contract seen).

**Lesson:** an archive copy was the expected fallback for a licence page, and it did not
exist. Pin the bytes when the source is first read.

## nonsa — third-party quotations and `<poem>` pages dropped (2026-10-06)

**Verdict** — kept as `release: None` (admission waits for the NERGAL pass,
polish-dynaword-v2 #60), 612 documents dropped: 12,304 → 11,692 documents,
16,883,883 → 15,439,683 tokens, 44,279,318 → 40,567,986 characters. Rebuilt from the
pinned wikiteam3 dump (`nonsa.pl-20251009-history.xml.zst`, sha1 99ed6dc7…, verified
by the script) with `src/build_nonsa_pl.py`; the retained rows equal the old shard
minus the dropped ones (all 8 columns identical for the 11,692 common ids). The drops
are in `third_party_drop`, applied after the QA and dedup gates; `nonsa.dropped.jsonl`
lists every dropped page with its reason:
- 287 pages in namespace 100 "Cytaty" (440,332 tokens): collections of quotes from
  named people, films, series and songs. Hand-checked 10 of 10 (seed "handcheck:").
- 325 pages whose wikitext has a `<poem` tag outside comments, `<nowiki>` and `<pre>`
  (1,003,868 tokens, 5.95% of the pre-drop total). The tag is a formatting tag, not a
  verse detector. Hand-checked 10: by my reading 4 are verse or lyrics (Jelcz,
  Dopalacze, Szczurem a prawdą, Das Kristallnacht), 6 are dialogue jokes, a game log,
  a forum quote or a line-by-line commentary on quoted text. The audit's "12 documents,
  0.5%" does not reproduce (the citations report called it not recountable).
  Dropping all 325 follows the audit literally; a per-page review of the list in
  `nonsa.dropped.jsonl` would keep part of them.
- Kept docs checked: 10 mentioning poem or quotes. The 2 kept pages containing `<poem`
  discuss the tag inside `<nowiki>` and are correctly kept.

**Reason** — the CC BY-SA 3.0 grant covers what contributors wrote, not the quoted
works. Verse, lyrics and quote collections reproduce third-party works (cf. wikiquote
and wikisource, withdrawn 2026-10-05).

### Not established

- Embedded quotes remain: 3,293 kept docs (28.2%) use `{{cytat}}`/`{{cytat2}}`/
  `{{cquote}}` (about 1.7% of wikitext characters in non-nested calls), 23 use
  `<blockquote>`. Plain-text lyrics are not detected. None reviewed.
- Nonźródła, Gra and Forum pages contain parody and quoted text; not reviewed.
- 99 of the 11,692 kept pageids return `missing` from nonsa.pl on 2026-10-06 (deleted
  since the 2025-10-09 dump; 1 more is now a redirect). The deletion logs give quality
  or relevance reasons, none for rights (89 community votes, 5 merges, 5 single
  deletions), and the rows are still in the shard.
- No `ns` column in the parquet; the namespace is in `nonsa.attribution.jsonl`.

References: polish-dynaword-v2 #60; `rights_audit_2026-10/REVIEW.html` (nonsa);
`rights_audit_2026-10/wikis/nonsa_ns.json`; `reports/citations/samorzad_nonsa.md`.
Licence read-only 2026-10-06: <https://nonsa.pl/wiki/Nonsensopedia:Prawa_autorskie>
(revid 1850937, 2026-03-02) states Creative Commons Attribution-Share Alike 3.0, and
`api.php?action=query&meta=siteinfo&siprop=rightsinfo` returns the same licence with
`http://creativecommons.org/licenses/by-sa/3.0/`.

**Lesson** — a tag name is not a content type. `<poem>` is used for jokes and logs as
well as verse, so measure what a rule drops before quoting its size.

## `edukacja_medialna_pl`: four lessons with embedded third-party quotations dropped

**Investigated:** 2026-10-06
**Verdict:** 252 → 248 lessons stay CC BY-SA 3.0 (728,987 → 713,052 tokens; the Michera pair is 7,282
tokens, the profile pair 8,653).
`src/clean_edukacja_medialna_pl.py` filters the SHA-pinned PR #26 Parquet and sidecars. Tracks the
2026-10 rights audit line "optionally remove the 3 quote passages".

### Verified

- Two lessons quote the film scholar Wojciech Michera by name, credited to a University of
  Warsaw course ("cytat z materiałów do kursu ..."): `montaz-materialu-filmowego` and
  `scenografia-charakteryzacja-kostiumy-i-aktorzy`, one passage each. Dropped whole.
- Two lessons (`kim-jestem-w-internecie`, `wizerunek-w-sieci`) share an invented-profile
  exercise with six "Ulubione cytaty" lines, each credited to a named person (Marilyn Monroe,
  Winston Churchill and others). Dropped whole as a fail-closed guess (see below).
- Kept on purpose: the attributed CC BY-SA Wikipedia excerpt (`ekonomiczne-aspekty-...`),
  the Constitution and ECHR article quotes (statutory text, Art. 4), lessons that only name
  films, books or point to a handout by URL (`uzaleznienia` links the Jakubik article, it does
  not embed it).

### Not established

- The audit says "2-3 short quotes" and "3 quote passages". Only the two Michera passages are
  reproducible. No third third-party quotation was found by reading every lesson that contains
  "cytat", "źródło" or a quote followed by an attribution. The profile exercise is a guess at
  what the audit counted. To keep those two lessons, delete the two `PROFILES` entries from
  `DROP` and run `git checkout 5ec3e18 -- data/edukacja_medialna_pl` first (+8,653 tokens).
- The raw XML is not in the checkout, so the filter works on the Parquet. The datasheet's
  "Reproduction" section still describes the original build.

### Side effects

- `src/test_edukacja_medialna_contribution.py` pinned the contributed data files by SHA-256
  through `artifacts/edukacja_medialna_pl_ontology_manifest.json`. It now compares the manifest
  pins with the input pins in the clean script, so the manifest stays untouched.

**Lesson:** an audit count ("3 quotes") is a lead, not a list. Reproduce it from the data before
writing the filter, and say which drops are confirmed.

## sejm_api_2011_2022 — provenance relabelled, cross-source dedup (2026-10-06)

**Verdict** — held (`release: None`). 118,902 → 32,460 documents, 135,072,403 →
30,766,807 tokens, 361,275,710 → 81,385,172 characters; parquet sha256
`8fad4b4eab069fb8a652f9dc6a2e7dbc8f49d1e1b6bfae0decd2b318d160a20a`. Admission waits for
the NERGAL pass (#60). Registry, card and NOTICE now say "Sejm API data via contributor
snapshot" and that `source_url` is a day-level listing, not a per-speech URL.

**Reason** — the source overlaps `parliamentary` (0.2.0) and `sejm_api` (0.2.5).
`src/clean_sejm_api_2011_2022.py` drops a speech when one document of a pinned reference
Parquet holds at least half of its 5-word shingles (containment >= 0.5; exact, no
MinHash). 86,442 speeches (73%) are dropped, 86,272 of them dated 2011-2019. By
reference set, with the maintainer's two #54 outcomes:

| Reference set | Dropped |
|---|---:|
| `parliamentary` + `sejm_api` (this build) | 86,442 |
| `parliamentary`, plenary / committee slice alone | 48,660 / 38,539 |
| `sejm_api` alone (outcome A, `parliamentary` withdrawn) | 216 |
| `sejm_api` + `parliamentary` without its committee slice (outcome B) | 48,833 |

Per-speech scores are in `sejm_api_2011_2022.decisions.jsonl`. If #54 changes
`parliamentary`, re-pin `REFERENCES` and rerun the script on the restored inputs.
Hand check of 20 pairs at 0.3-0.7: 14 carry the speech's own wording in the reference
(4 of 10 under 0.5, 10 of 10 from 0.5); the 6 that do not are dated 2020 or later.

### Not established

- Outcome B is the audit's `ppc_classify` heuristic, which is unvalidated: 10 of the 11
  committee-labelled references in the hand-check sample are plenary sittings.
- Containment is against single reference documents; the union of several is not scored.
- No minimum size: 1,236 drops have under 10 shingles and 1,177 have 10-19. Of the 170
  drops dated 2020 or later, 12 sampled at random were 9 stock phrases or stage
  directions, 2 report templates and 1 real reuse, so about 100 may be false drops.
  65 of the 170 have 50 or more shingles; two were read (reuse of earlier text), the
  rest are unchecked. A minimum-shingles or date guard is the likely fix.
- Agenda titles without a number ("Oświadczenia.") are not stripped from the snapshot
  headers before scoring; numbered titles end correctly in 98.7% of 14,090 checked.
- Whether the snapshot matches the live API (20-row spot check) and whether the
  www.sejm.gov.pl terms reach API output; the API documents no licence.
- `src/fetch_sejm_api_2011_2022.py` is unchanged because the ontology manifest pins its
  hash; rebuilding with it brings back the old wording and the 118,902 rows.

References: polish-dynaword-v2 #57, #54, #60; `rights_audit_2026-10/REVIEW.html`
`#sejm_api_2011_2022`.

**Lesson** — "contained in a released source" needs a size floor and a date check: a
containment score of one stock sentence says nothing about reuse.

## ted_pl: two licence bases cited, text rebuilt from the TED API with fixed PII patterns (2026-10-06)

Audit actions from slayerlabs/polish-dynaword-v2#57; stays `release: None` until the
NERGAL pass (#60).

### Verified

- TED legal notice (read 2026-10-05, sha256 `c010943c01af22cf9cb30a32ca4044a3a1ee4e36a090e2dd2d8ee3d15a498997`):
  procurement notices "can be freely reused, for commercial or non-commercial purposes"
  unless otherwise noted. The CC BY 4.0 sentence covers SIMAP editorial content and the CC0
  sentence covers system metadata, not the notice text.
- Decision 2011/833/EU is dated 12 December 2011 (the old datasheet said 21 December) and covers
  Commission documents (arts 1, 2(1)); it is not shown to license notices written by national authorities.
- Full scan of 15,544 docs: 13 e-mails in 13 docs and 480 phone numbers had survived the old
  rules. Of 33 docs containing "@", 20 are not addresses. Fixed in `redact_pii`.
- Rebuild: the 15,544 shipped notice ids were re-fetched from the TED Search API (XML
  fallback), 0 unresolved, raw bytes hash-identical to the original acquisition, no key, terms
  or throttling. `redact_pii` is applied once to the raw text (`build_ted_pl.py rebuild`, tests in
  `src/test_ted_pl_contract.py`); the earlier `rescrub` step is gone. Tokens 42,475,098
  (v0.2.5) to 42,471,550 (rescrub) to 42,473,050; `[Telefon]` 470 to 950 to 785 spans; `[PII]`
  1,080 to 1,093 to 1,093 spans.
- 164 `[Telefon]` spans in 81 docs from the rescrub are restored (amounts, quantities, part
  numbers, a procurement code, a bank account); 10 read in context, 10 of 10 correct. A fresh
  sample of 12 `[Telefon]` spans: 11 phones, 1 part number.

### Not established

- Polish basis (Copyright Act art. 4 pkt 2, Open Data Act arts 2 pkt 8, 3, 5, 14, 15) is Inferred:
  art. 5 covers an entity's own BIP/system, TED is an EU system; about 20% of authors carry a
  company form and may not be "podmioty zobowiązane". No lawyer sign-off.
- Personal names are not redacted (heuristic lower bound: 172 docs); NERGAL handles them.
- Labelled PESEL/NIP/REGON are still replaced by `[PII]` for companies too (pattern unchanged
  since v0.2.5), so company identifiers did not come back. 242 raw matches (NIP in 66 docs,
  REGON in 38). One contractor looks like a natural person with a NIP next to the name
  (`ted_pl_167339-2024`), three more are partnerships with personal names; no author is a person.
  Bank accounts (1,111 docs) and KRS numbers are not redacted.
- Dotted 3-3-3 part numbers are still read as phones (5 spans in 2 docs).

**Lesson:** a label-anchored phone regex with no number-shape guard turns amounts and part
numbers into "phones"; check redactions for false positives, not only misses. Keep the raw
download recipe runnable: a rescrub of already-redacted text cannot undo a false positive.

## wikinews — Polish Wikinews (rebuild from the official dump, 2.5/4.0 labels)

**Investigated** — 2026-10-06 (October 2026 rights audit, `artifacts/rights_audit_2026-10/reports/wikis_ugc.md`,
`reports/citations.md` L17-26; slayerlabs/polish-dynaword-v2#57).

**Verdict** — `kept`, relabelled; rebuilt from the official dump so the label follows the
page's last revision. Documents, characters and tokens are unchanged from v0.2.5
(24,386 / 32,524,646 / 12,141,355); the text multiset is identical. Labels per record:
`CC-BY-2.5` 17,581 documents (8,956,397 tokens, 73.8%), `CC-BY-4.0` 6,805 (3,184,958, 26.2%);
v0.2.5 labelled every row `CC-BY-2.5`. Row ids are now `wikinews_<page id>`; `author` is the
page URL and `created` the last-revision date (v0.2.5: empty `author`, and the same range
string "2005-01-01, 2025-12-31" as `created` on every row). The parquet `added` column is
2026-07-19 (v0.2.5: 2026-06-15).

**Enumeration** — `plwikinews-20260601-pages-articles.xml.bz2` (sha1 `bfad5f33…` matched to
the dump's `sha1sums.txt`), parsed by `src/fetch_wikimedia.py`: 65,253 pages → 47,845 in
namespace 0 → 23,366 redirects and 87 pages under 200 characters skipped → 24,392 written →
6 non-Polish dropped by `build_dynaword.py` → 24,386. Text conversion is the v0.2.5 one
(plain `strip_code`): `<ref>` bodies, `Kategoria:` links and `thumb|` file options stay in
the text; nothing was changed in this PR. One sidecar line per row
(`wikinews.attribution.jsonl`).

**License evidence** — verified, read 2026-10-06: pl.wikinews terms of use (oldid 342796:
CC BY 2.5 until 2024-12-16, CC BY 4.0 since), Gerrit 1101867 (merged 2024-12-16 08:07 UTC),
Phabricator T381421, the 2.5 legal code (§4(a): no later-version clause), pl.wikinews
`rightsinfo` (BY 4.0). The 08:32:24 UTC sync time is from the audit and was not re-read.
Cutoff is 2024-12-16 00:00 UTC for the whole day: 6 documents were last edited that day, 2 of
them before 08:32:24 UTC and so strictly under 2.5. Inferred: because CC BY 2.5 has no
later-version clause, the `CC-BY-4.0` label for post-switch pages rests on the site's terms
page alone. This is a question for the lawyer, separate from E1 (wiki 3.0/4.0 labelling,
v2#56), whose answer is still pending. The terms page also says articles created before
2005-09-25 are public domain; 0 released documents were last edited before that date, so the
clause does not change a label here. Evidence in `data/wikinews/wikinews.license-evidence.json`.

**Blockers** — none to merge. Open: lawyer on the 2.5 → 4.0 label and on E1; the 2 boundary
documents; copyright pastes in news text (the audit's 100-document press-copying sample was
not run); the markup residue above.

## eurlex — relabelled to the EU reuse policy, 92 IASB documents dropped (2026-10-06)

**Investigated** — 2026-10-06 (October 2026 rights audit, polish-dynaword-v2 #57; lawyer
question #56 B5; `rights_audit_2026-10/reports/citations.md`).

**Verdict** — kept and relabelled, the 92 IASB-notice documents dropped (maintainer
decision 2026-10-06): 243,060 -> 242,968 documents, 5,976,949,249 -> 5,902,889,135
characters, 2,378,055,718 -> 2,350,709,056 tokens. `CC-BY-4.0` becomes
`LicenseRef-EU-Reuse-Policy` in the per-row `license` column, the registry and the card,
by `src/clean_eurlex.py` (tested in `src/test_eurlex_contract.py`; every kept row is
otherwise identical to v0.2.5). The drop is pinned to the 92 hand-checked ids and fails
closed if the IASB rule and the list disagree. The Polish Copyright Act Art. 4 basis is
removed from the card.

**License evidence** — verified, read 2026-10-06: the EUR-Lex legal notice
(<https://eur-lex.europa.eu/content/legal-notice/legal-notice.html>, sha256 `7994b96f…`)
bases the Commission's reuse policy on Decision 2011/833/EU and allows re-use of the legal
documents published in EUR-Lex "unless otherwise specified". Its CC BY 4.0 grant covers
only the website's editorial content, summaries of EU legislation and consolidated texts,
so a corpus-wide CC-BY-4.0 label overstated the grant. Decision 2011/833/EU
(<https://eur-lex.europa.eu/eli/dec/2011/833/oj/eng>, "In force", sha256 `1e867b79…`):
Art. 4 reuse for commercial and non-commercial purposes, Art. 6(2) source
acknowledgement and no distortion, Art. 2(2)(a)-(b) and recital 14 exclusions. Inferred:
the Decision covers Commission documents, so acts of other institutions (Council,
Parliament, ECB, EESC, CJEU) rest on the notice's general sentence. Evidence in
`data/eurlex/eurlex.license-evidence.json`.

**Third-party check** — `python3 src/clean_eurlex.py --scan` on the v0.2.5 input:
- 92 documents (74.1M characters, 1.24%) contain the IASB notice: reproduction allowed in
  the EEA, all rights reserved outside it. They are 64 OJ acts adopting or amending
  IAS/IFRS and 28 consolidated versions (20 of Regulation 1126/2008). The notice is the
  "special conditions of use" the legal notice names for the International Accounting
  Standards. All 92 matched lines and all 64 titles hand-checked. Two more documents have
  a standard title line without the notice (67,283 characters) and were not counted.
  Dropped. None of the 92 is in the (c)-sign or header counts below, so those are unchanged.
- 97 documents contain a (c) sign; 33 are third-party credits or short data citations
  (photographs, maps, OECD, IHS Markit, McKinsey, Springer), 25 EU credits, 7 binary
  garbage, 32 glyph noise or list markers. Labelled by reading the first occurrence in
  each; the credited images are not in the text.
- Header patterns: ECB acts 1,414 documents (1.00% of characters), EESC 5,154 (2.16%),
  state aid `SA.nnnnn` 2,941 (1.26%), UN ECE regulations 253 (0.39%). These are EU-body or
  Commission texts, not third-party under the notice; they matter only if those bodies'
  reuse terms differ. The audit's rough shares (ECB ~4%, state aid ~2.5%) do not
  reproduce with these patterns.
- No large-scale reproduction of non-EU copyrighted works was found. Recall for third-party
  text the patterns do not target is unknown.

**Blockers** — none to merge. The notice reserves rights outside the EEA and the Hub is not
EEA-limited, so the 92 documents were dropped; the (c)-sign and header documents are kept.
Ids are in `data/eurlex/eurlex.third-party-scan.json`.

**Not established** — SpeakLeash's terms for `eurlex_corpus`; whether the ECB, EESC,
Council, Parliament and the Court of Justice publish reuse terms of their own; the
Publications Office's position on non-Commission documents (lawyer question, #56 B5).

**Lesson** — a reuse policy is not a Creative Commons licence: copy the clause, not the
label, and read what the grant covers before using its name.

## wikibooks — Polish Wikibooks (rebuild from the official dump, 3.0/4.0 labels, GFDL review)

**Investigated** — 2026-10-06 (October 2026 rights audit, `artifacts/rights_audit_2026-10/reports/wikis_ugc.md`,
`reports/citations.md` L17-26; slayerlabs/polish-dynaword-v2#57).

**Verdict** — `kept`, relabelled; rebuilt from the official dump so the label follows the
page's last revision. Documents, characters and tokens are unchanged from v0.2.5
(9,112 / 38,892,589 / 15,571,295); the text multiset is identical. Labels per record:
`CC-BY-SA-3.0` 4,461 documents (6,102,574 tokens, 39.2%), `CC-BY-SA-4.0` 4,651 (9,468,721,
60.8%); v0.2.5 labelled every row 3.0. Row ids are now `wikibooks_<page id>`
("Przewodnik po instalacji Debiana 3.0": `wikibooks_2419570` in v0.2.5, `wikibooks_1710` now);
`author` is the page URL and `created` the last-revision date (v0.2.5: empty `author`, and
the same range string "2004-01-01, 2025-12-31" as `created` on every row). The parquet
`added` column is 2026-07-19 (v0.2.5: 2026-06-15).

**Enumeration** — `plwikibooks-20260601-pages-articles.xml.bz2` (sha1 `198ddcbe…` matched to
the dump's `sha1sums.txt`), parsed by `src/fetch_wikimedia.py`: 34,967 pages → 12,838 in
namespace 0 → 1,925 redirects, 1,247 pages under 200 characters and 23 exact duplicates
skipped → 9,643 written → 531 non-Polish dropped by `build_dynaword.py` → 9,112. Text
conversion is the v0.2.5 one (plain `strip_code`): `<ref>` bodies, `Kategoria:` links and
`thumb|` file options stay in the text; nothing was changed in this PR. One sidecar line per
row (`wikibooks.attribution.jsonl`).

**License evidence** — as for wikivoyage: Terms of Use oldid 554823 / 98524, the 3.0 and 4.0
legal code, pl.wikibooks `rightsinfo` (BY-SA 4.0), read 2026-10-06. Cutoff 2023-06-07 00:00
UTC; 2 documents were last edited that day. E1 (v2#56) is still pending.

**GFDL** — 435 documents (259,065 tokens, 1.7%) were last edited before 2009-06-15 (the
start of CC BY-SA on the Wikimedia sites, per the audit); they carry `CC-BY-SA-3.0`, counted
and not acted on. The audit's
"24 GFDL-mentioning documents" cannot be reproduced (its pattern is not recorded; the closest,
`\bFDL\b|Wolnej\s+Dokumentacji`, gives 25 documents / 69,072 tokens and neither document the
audit names). A wider case-insensitive pattern matches 45 documents (122,192 tokens, 0.78%);
all 45 were read and sorted in `data/wikibooks/wikibooks.gfdl-review.json`: 7 state that their
own text is under the GNU FDL (42,834 tokens), 10 name a GFDL-licensed upstream work (6,903),
13 link the licence page only, 13 discuss licences, 2 match by accident. The largest is
`wikibooks_30538` (35,619 tokens). No label and no drop follows from the review; the 17
own-text and upstream documents (49,737 tokens) are the ones where a GFDL-only grant could
matter, and keeping them is left to the maintainers and the lawyer.

**Blockers** — none to merge. Open: E1; the 17 documents above. Two pages seen during the
review and not checked: `wikibooks_27716` (145 tokens) lists third-party course sources with
their licences, among them CC-BY-NC 3.0 and "All Rights Reserved"; `wikibooks_27071` (1,287
tokens) says it was moved in 2013 from the "Wolne podręczniki" portal with its creators'
consent and does not state the original licence. The markup residue above also stays.

## wikibooks — 46 pages withheld (update to the section above)

**Investigated** — 2026-10-06 (copyright sample and licence trace of the follow-up, then the
maintainer's decision; slayerlabs/polish-dynaword-v2#57).

**Verdict** — `kept`, 46 pages withheld. Maintainer approval 2026-10-06, relayed by the
coordinator. The numbers in the section above (9,112 documents, 38,892,589 characters,
15,571,295 tokens; 3.0: 4,461 / 6,102,574 tokens; 4.0: 4,651 / 9,468,721) are the v0.2.5 and
pre-drop values; the release now holds 9,066 documents, 38,195,523 characters and 15,313,229
tokens (3.0: 4,425 / 5,916,490; 4.0: 4,641 / 9,396,739). Withheld: 46 documents, 697,066
characters, 258,066 tokens (36 labelled 3.0, 10 labelled 4.0).

**Withheld** — `wikibooks_23933` (Historia powszechna/Rzeczpospolita pod panowaniem królów
elekcyjnych, 10,753 tokens): copy of sciaga.pl, Earwig 0.964, 43.9% of the words, the source
captured since 2007-10-11, the wiki page created 2011-03-28. `wikibooks_27715` (Od lamera do
hakera/Linux, 3,761 tokens): Polish translation (Paweł Niewiadomski, linuxpl.org) of chapter
6 of The Linux System Administrators' Guide (Lars Wirzenius, Joanna Oja); the page's own
closing note gives CC BY-NC 3.0. `wikibooks_27602` (Od lamera do hakera/Procesor, 7,548
tokens): text pasted on 2013-06-25, the source link lost, no licence given; Earwig suspects
slideshare.net/slideshow/mikroprocesory/32678654 (0.896). `wikibooks_26962`
(Mentalizm/Wprowadzenie, 1,313 tokens) and the 42 other pages of the Mentalizm book
(234,691 tokens; hub `wikibooks_26961` included): the introduction states non-commercial
share-alike terms (NC wording removed 2013-01-13, restored 2015-02-02); the other pages carry
no notice of their own and are withheld with the book. Together 236,004 tokens.

**Mechanism** — the ids and reasons live in `DROPPED_PAGES["wikibooks"]` in `src/rights.py`;
`fetch_wikimedia.py prepare` skips them and fails if one is not reached. Funnel: 34,967 pages
→ 12,838 in namespace 0 → 1,925 redirects, 1,247 short pages, 23 exact duplicates and the 46
withheld pages skipped → 9,597 written → 531 non-Polish dropped → 9,066. Diff against the
previous parquet: only the 46 ids removed, every other row identical. The 2 documents of the
boundary day and the 435 GFDL-era documents are unchanged. None of the 46 is among the 45
documents of `wikibooks.gfdl-review.json`; only its `source_tokens` (15,313,229) and the share
in its summary (0.80%) were regenerated.

**Blockers** — none. Open: the copyright sample read 79 of the 9,112 pages and 90 more by one
contributor; `wikibooks_27716` (145 tokens, lists CC-BY-NC and all-rights-reserved sources by
name; kept) and `wikibooks_27071` (1,287 tokens, original licence not stated; kept);
the 25 released pages that the "Od lamera do hakera" hub links in other books were not
traced; E1 (v2#56); the 2 boundary documents; the 17 own-text and upstream GFDL documents;
the markup residue.

## wikivoyage — Polish Wikivoyage (rebuild from the official dump, 3.0/4.0 labels)

**Investigated** — 2026-10-06 (October 2026 rights audit, `artifacts/rights_audit_2026-10/reports/wikis_ugc.md`,
`reports/citations.md` L17-26; slayerlabs/polish-dynaword-v2#57).

**Verdict** — `kept`, relabelled; rebuilt from the official dump so the label follows the
page's last revision. Documents, characters and tokens are unchanged from v0.2.5
(13,645 / 45,184,300 / 17,128,200); the text multiset is identical. Labels per record:
`CC-BY-SA-3.0` 9,041 documents (4,233,610 tokens, 24.7%), `CC-BY-SA-4.0` 4,604 (12,894,590,
75.3%); v0.2.5 labelled every row 3.0. Row ids are now `wikivoyage_<page id>`; `author` is
the page URL and `created` the last-revision date (v0.2.5: empty `author`, and the same range
string "2012-01-01, 2025-12-31" as `created` on every row). The parquet `added` column is
2026-07-19 (v0.2.5: 2026-06-15).

**Enumeration** — `plwikivoyage-20260601-pages-articles.xml.bz2` (sha1 `1a790d37…` matched to
the dump's `sha1sums.txt`), parsed by `src/fetch_wikimedia.py`: 15,468 pages → 14,321 in
namespace 0 → 630 redirects, 36 pages under 200 characters and 1 exact duplicate skipped →
13,654 written → 9 non-Polish dropped by `build_dynaword.py` → 13,645. Text conversion is the
v0.2.5 one (plain `strip_code`): `<ref>` bodies, `Kategoria:` links and `thumb|` file options
stay in the text; nothing was changed in this PR. One sidecar line per row
(`wikivoyage.attribution.jsonl`).

**License evidence** — verified, read 2026-10-06: Wikimedia Terms of Use oldid 554823 (CC BY-SA
4.0 and GFDL, in effect June 7, 2023; sha256 `c96ef93b…`), oldid 98524 (CC BY-SA 3.0, June 16,
2014), the 3.0 and 4.0 legal code, pl.wikivoyage `rightsinfo` (BY-SA 4.0). Inferred: a page
edited after the switch may carry earlier 3.0 text under 4.0 (3.0 §4(b)(ii)). The cutoff is
2023-06-07 00:00 UTC, not the audit's "06-08 if time of day unknown": 06-08 would put that
day's 4.0 edits under 3.0, which 4.0 §3(b)(1) does not allow; 6 documents were last edited on
2023-06-07 and may carry 3.0 text under a 4.0 label. The footer switch date was not checked.
The answer to E1 (v2#56) is still pending. GFDL era (last edit before 2009-06-15): 0
documents. Evidence in `data/wikivoyage/wikivoyage.license-evidence.json`.

**Blockers** — none to merge. Open: E1; the 6 boundary documents; copyright pastes (travel
text copied from guidebooks and tourist sites was not sampled for Wikivoyage); the markup
residue above.

## wikivoyage — two pages withheld (update to the section above)

**Investigated** — 2026-10-06 (copyright sample of the follow-up, then the maintainer's
decision; slayerlabs/polish-dynaword-v2#57).

**Verdict** — `kept`, two pages withheld: `wikivoyage_1219` (Barcelona/Sagrada Família) and
`wikivoyage_910` (Królewiec). Maintainer approval 2026-10-06, relayed by the coordinator. The
numbers in the section above (13,645 documents, 45,184,300 characters, 17,128,200 tokens;
3.0: 9,041 / 4,233,610 tokens; 4.0: 4,604 / 12,894,590) are the v0.2.5 and pre-drop values;
the release now holds 13,643 documents, 45,176,480 characters and 17,125,240 tokens (3.0:
9,040 / 4,232,587; 4.0: 4,603 / 12,892,653). Withheld: 1219 is 3.0, 1,023 tokens, 2,771
characters; 910 is 4.0, 1,937 tokens, 5,049 characters.

**Reasons** — 1219: edited from barcelona.przewodnik-turystyczny.pl (the page's own
bibliography names it); 432 of 2,782 characters shared verbatim (Earwig 0.23), a Wayback
capture of 2006-06-16 already holds the passages, the wiki page was created 2009-06-04. 910:
2,021 of 5,006 characters shared with kaliningraddotblog.wordpress.com/2018/05/07/krolewiec-konigsberg/
(Earwig 0.78); the blog post is dated 2018-05-07, the text was added to the wiki on
2021-02-23 (revision 108903).

**Mechanism** — the ids and reasons live in `DROPPED_PAGES["wikivoyage"]` in `src/rights.py`;
`fetch_wikimedia.py prepare` skips them and fails if one is not reached. Funnel: 15,468 pages
→ 14,321 in namespace 0 → 630 redirects, 36 short pages, 1 exact duplicate and the 2 withheld
pages skipped → 13,652 written → 9 non-Polish dropped → 13,643. Diff against the previous
parquet: only the two ids removed, every other row identical; the 6 documents of the boundary
day and the 0 GFDL-era documents are unchanged.

**Blockers** — none. Open: the copyright sample read 80 of the 13,645 pages (40 per label),
so 13,565 were not checked; two further pages stay in the release with the source's direction
unresolved (`wikivoyage_2477`, `wikivoyage_989`); E1 (v2#56); the 6 boundary documents; the
markup residue.

## `studia_bas`: licence version resolved to CC BY 3.0 PL

**Investigated:** 2026-10-06
**Verdict:** no data change; 112 documents stay CC BY 3.0 PL, no per-row version label. The
resolution is recorded in `data/studia_bas/studia_bas.license-evidence.json`
(`version_resolution`) and pinned by `src/test_studia_bas_contract.py`. Tracks the 2026-10
rights audit action "check live site manually (CAPTCHA); resolve 3.0 vs 4.0".

### Verified

- The journal's own pages say CC BY 3.0 PL: seven Wayback captures (2021-07-11 to 2024-09-08,
  open-access page and author instructions, PL and EN) and the three live pages on
  2026-10-06. The cutoff is issue 2(66)/2021 wherever one is stated, which is where the data
  starts. The newest issue listed live is 2(82)2025, the last one in the data.
- Plain GET requests with an identifying User-Agent returned the live pages, not the
  human-verification page, so nothing was answered or bypassed. Archived captures from
  2025-11-15 on are the verification page, and it may come back.
- DOAJ is the only source that says 4.0 (journal-level `license.url`, record updated
  2026-09-17). Its own `copyright.url` and `oa_statement` point at the journal pages above.
  Crossref holds no licence deposit for any of the 246 works.
- Live PL author-instructions paragraph now names "Studia i Ekspertyzy Sejmowe" and drops the
  "Od numeru 2(66) 2021" cutoff (2022-2024 captures have both). It still says CC BY 3.0 PL.

### Not established

- The notice inside each PDF (the fetcher's gate and the evidence file's PDF quotes) was not
  re-read: that needs the 17 issue PDFs (51,701,823 bytes) from orka.sejm.gov.pl.
- The journal's wording between 2024-09-08 and 2026-10-06 (no readable capture). By the
  four-a-year numbering, issues 80-82 (11 rows) fall in that window.
- CEJSH and BazHum: guessed record URLs returned 404.

**Lesson:** the audit found the live site behind a verification page, yet a plain read-only
GET returned it on 2026-10-06. Retry that before routing a check to a human.

## biblioteka_nauki — licence versions restored from the API (2026-10-06)

**Verdict** — `added`, kept for v0.2.6 minus 83 rows (polish-dynaword-v2#57; maintainer approval
2026-10-06). 42,068 → 41,985 documents, 673,112,676 → 669,909,430 tokens (−3,203,246),
1,747,387,478 → 1,739,004,913 characters. The `license` column now carries the version the API
gives per record: CC-BY-4.0 17,768, CC-BY-3.0-PL 8,098, CC-BY-3.0 2,413, CC-BY-2.0 1, CC-BY-SA-4.0
10,944, CC-BY-SA-3.0-PL 2,754, CC-BY-SA-3.0 7. Before, 28,335 rows said `cc-by` and 13,733
`cc-by-sa`, and the card said 4.0 for all. Only that column changed: same ids in the same
order, same text hashes for the 41,985 shipped rows (digest in `stats.json` and the evidence file).

**Method** — `data/biblioteka_nauki/recover_license_versions.py` asked
`bibliotekanauki.pl/api/{type}/{id}` once per row of the previous release (2026-10-05/06, one
request clock, 3 workers, back-off on 429/5xx, stop on 403). 42,056 answered 200; no block, no
CAPTCHA. The responses are committed as `biblioteka_nauki.license-api.json`. The cleaner labels
a row only when the `creativecommons.org` URL, the English name and the icon agree with each other
and with the earlier label, for CC BY / BY-SA 2.0 to 4.0 (generic or PL); anything else is withheld.

**Withheld (83)** — 12 rows the API now answers with 404 (`articles_512191`, `604103`, `605338`,
`694689`, `699376`, `699392`, `699396`, `699422`, `699424`, `699432`, `699440`, `699442`; asked
again by hand, 404 again); 56 rows whose text prints an NC or ND licence (Krytyka Prawa 2019 21,
Journal of Health Sciences 2014 12, 23 others) and 15 rows whose text says it is reprinted or
translated by permission of a publisher (Oxford University Press 3, others 12), both lists decided
by the maintainer on 2026-10-06 and kept by id, with a reason per id, in
`src/clean_biblioteka_nauki.py`. `articles_2011995`, whose API licence is CC BY-SA 4.0 while the
earlier recovery said CC BY, is kept as CC-BY-SA-4.0. Why the platform dropped the 12 records, and
whether it changed that licence after the first recovery, is not known.

**Not resolved** — the 3 Public Domain Mark rows were already gone (Hub PR 62). The 128 rows
the earlier review found with "all rights reserved" in the text could not be reproduced exactly (regex variants give
121 to 137, 127 for the whitespace-tolerant one; 4 of the 127 are among the withheld rows). All of them have a CC BY or CC BY-SA licence in
the API, none is all-rights-reserved there. They stay in; the footer-versus-API conflict is a
rights decision. 5 kept rows carry `reprintPublication` (2) or `translationPublication` (3);
none has `restricted`. The card and registry say the text came from the SpeakLeash
redistribution (Hub PR 11) with licence and attribution recovered per document from the API; the
text-assembly script (`build_parquet.py`) is not in the repo or on the Hub, so how the text layer
was assembled is not re-verified.

**Lesson** — a label that drops the version is not a licence. 28,280 rows are CC BY and
13,705 CC BY-SA, and 13,273 of the 41,985 rows are not 4.0.

## sejm_api — "Przebieg posiedzenia" trailers cut, appended statements split out (2026-10-06)

**Finding** (slayerlabs/polish-dynaword-v2#57) — the PiotrSty snapshot appends to 34,980
of 38,812 speeches the header of a later statement of the same sitting day
("Przebieg posiedzenia {name}: {term}. kadencja, {n}. posiedzenie, {d}. dzień - {label}
… ({dd-mm-yyyy})"), and after 2,466 of those headers that statement's text. The trailers
are 23,485,670 characters (24% of the source), released in v0.2.5 (Hub PR 16).

**Fix** — `src/clean_sejm_api.py` (inputs pinned by SHA-256 to the pre-cut files at
`262df1b`, unchanged since v0.2.5) cuts every trailer with `sejm_api_common.split_trailers` and emits each
appended statement as a row of its own (`split_release_rows`): speaker = the name the
header gives ("Przebieg posiedzenia {name}:"), date and term of the parent row,
`split_from` = parent id, `has_events: null`. Rows under 200 characters after the cut
(1,318 parents, 53 split rows) and 8 new exact duplicates are dropped. 38,812 → 39,899
documents, 36,335,587 → 30,860,439 tokens. Same-speaker follow-ups (373) are kept as
separate rows; agenda titles ("N. punkt porządku dziennego: …", "Oświadczenia.") before a
follow-up are dropped.

**Author of a split row** — the first push of this PR credited split rows in the snapshot's own
`speaker` convention, the header label minus its first word. That is right for "Poseł Jan
Kowalski" but leaves an office fragment for officials ("Podsekretarz Stanu w Ministerstwie
Zdrowia Janusz Cieszyński" -> "Stanu w Ministerstwie Zdrowia Janusz Cieszyński"). The header's
name fixes 2,052 published split rows; 129 more lose a "(tekst niewygłoszony)" suffix and 12 a
"Sprawozdawca" prefix that the snapshot keeps on its own rows. Text, counts and stats are
unchanged; the ids of these 2,193 rows change, since the id hashes the speaker.

**Upstream quirk** — the snapshot labels the 11th sitting of term 10 (2024-05-08/09/15)
"11. kadencja" in 696 headers. The split keeps the parent's term; a trailer dated on
another day than its parent fails closed (none occur).

**Also fixed** — `fetch_sejm_api.py` read the registry end date as the latest date in
the entry, which picked up `added` (2026-07-02) once entries carried it, and a rerun
then failed temporal novelty. It now parses the `created` range only, checks against
`parliamentary` alone (`parlamint_pl` left in Hub PR 54 and never shipped), and keeps an
existing `sejm_api` registry entry that meets the contract instead of failing.

**Lesson** — check a scraped snapshot for navigation text glued to the record: a trailer
in 90% of rows hid in plain sight because each row still began with the right speech.

## sejm_api_2011_2022 — trailers cut before the cross-source dedup (2026-10-06)

**Finding** (slayerlabs/polish-dynaword-v2#57) — the same "Przebieg posiedzenia" trailers
as in `sejm_api` (section above) are in 105,920 of the 118,902 pre-dedup speeches:
97,723,744 characters. Hub PR 76 only stripped them from the shingled text when it scored
containment; the published rows kept them.

**Fix** — `src/clean_sejm_api_2011_2022.py` now runs `sejm_api_common.split_release_rows`
on the pinned pre-dedup inputs before the containment search, so the published rows lose
their trailers and each appended statement is scored and published on its own (15,648
split rows; 5,374 cut speeches and 2,189 split rows under 200 characters and 20 new exact
duplicates dropped; 126,967 rows enter the search). The `sejm_api` reference is re-pinned
to its own post-cut Parquet (`ba2daa20…`). Header stripping in the shingler is gone: after
the cut no header is left in the inputs or the references. 118,902 → 33,106 documents
(Hub PR 76: 32,460), 135,072,403 → 26,112,679 tokens (30,766,807), 93,861 rows contained
(86,442), 93,784 of them dated 2011-2019. Of the 32,460 rows Hub PR 76 kept, 1,882 fall
under 200 characters once their trailer is cut (504 of them dated 2011-2019), 4 now reach
0.5 and 5 drop below it; 2,527 split rows are added, 180 of them from a speech that is
itself dropped. Kept 2011-2019 rows go from 1,697 to 1,281. Parquet sha256
`622228ca4c287f91792b7da3fb97c098610c1cf890f1bdc08922dc3aca3742db`.

**Not established** — the card's hand checks (12 of the post-2019 drops, 20 pairs at
0.3-0.7) were made on the Hub PR 76 build and were not repeated.

## sejm_interpellations — licence evidence and SIS caveat (2026-10-06)

**Finding** (slayerlabs/polish-dynaword-v2#57) — the source had no license-evidence.json. The
Regulamin Sejmu places interpellations, replies and written questions in the System Informacyjny
Sejmu (arts. 192 ust. 7, 193 ust. 2, 195 ust. 2), whose terms page asks for a source note and bars
SIS material from commercial offers; nothing on api.sejm.gov.pl says its output is part of SIS.
Replies are written by ministries. Questions quote third-party text (rights audit sample of 3,900:
2.2% quote 400+ characters). The registry `added` was the 2026-07-19 fallback stamp, older than
the data (last record 2026-09-03).

**Fix** — `sejm_interpellations.license-evidence.json` records both statutory bases with file
hashes, the SIS wording (Wayback 2026-06-03), the placement articles, the reading that the
commercial clause is not an Open Data Act art. 15 ust. 1 condition, and the third-party sample.
The datasheet states the caveats and gives the requested source note; the registry `added` is
2026-09-05 and the entry has a `legal_note`. The Parquet rows still carry the global `ADDED`
(2026-07-19) until `build_dynaword.py` stamps the registry date (v2#80).

**Not established** — a lawyer's reading of the SIS terms; the SIS page on the fetch date (bot
challenge); the Kancelaria Sejmu BIP reuse page; personal names (NERGAL, v2#60).

## sejm_committee_transcripts — only statements by official speakers (2026-10-07)

**Finding** (October 2026 rights audit) — art. 4 pkt 2 of the copyright act covers official
documents, and a committee transcript also holds statements by guests, experts and
representatives of unions, NGOs and companies, and about 2,900 anonymous "Głos z sali" turns.
The audit's heuristic put the non-official speakers at about 5% of the characters. The first
allowlist cut 11.9%, but it split turns only at lines that start with a role word, so a guest's
label (`Świadek`, `Członek zarządu ...`) and speech stayed inside the previous official turn
(`10_SPC_62_14`, `10_SKGK_3_24`): the real share of non-official text was higher, and the shard
still carried it. Hub PR 37 added all 653,411 turns to main.

**Fix** — `fetch_sejm_committee_transcripts.SPEAKER_RULES` is a fail-closed allowlist: a piece
is kept only if its `author` label, NFC-normalised, fully matches one anchored rule (MPs, chairs
who are MPs, Marshals of the Sejm and Senate, senators, members of the government, secretaries and
undersecretaries of state in a ministry or the KPRM, heads of bodies named in the Constitution,
Biuro Legislacyjne legislators, BAS/BEOS experts and specialists, Sejm committee secretaries,
heads of the Sejm and Senate offices). One label-line detector (`is_label`/`split_at_labels`) and
one page-block stripper (`strip_page_blocks`) live in the fetch script and are used by `parse_pdf()`
and by `src/clean_sejm_committee_transcripts.py`, which re-filters the pinned pre-filter release
(the PDF cache is not kept; SHA-256 pins, input stats checked against the input Parquet): every
stored turn is cut at each label line (piece 0 keeps the id and author, later pieces get the label
as author and the id `<id>_<k>`), page blocks (page number + stenographer initials, `Pełny Zapis
Przebiegu Posiedzenia:` title, `Komisji ... (Nr N)`) are removed as a unit with a word hyphenated
across one rejoined (v2#95), and every piece is classified. 653,411 turns → 740,018 pieces →
613,722 records (727,570 raw + 86,607 split out; joint-sitting duplicates 74,117, exact duplicates
77, under 15 characters 16,080, not official 110,181), 443,928,915 → 298,555,500 characters,
166,310,754 → 112,003,155 tokens; the 110,181 dropped pieces are 127,167,730 characters (28.6%
of the unfiltered release) under 21,364 labels. 122,521 page blocks removed, 4,929 hyphenated
words rejoined. Of the 3,790 labels kept, `mp` 220,125 pieces, `mp_committee_chair` 302,535,
`state_secretary` 45,800, `sejm_legislator` 34,563, `sejm_committee_secretary` 1,873,
`government_member` 4,154, `constitutional_organ_head` 841, `sejm_senate_marshal` 1,490,
`sejm_bureau_expert` 1,048, `sejm_senate_office_head` 720, `senator` 573.
`sejm_committee_transcripts.speaker-roles.csv` lists every role label (the label without the
person's name) with its decision and volume. Parquet sha256
`8c7d038e83bb32afd465271198c6ea673069647dc28e21eee9b3fca56c885118`, attribution
`9ff1c0c39b2c4b9e81f1ef63f8c33b11e7a4fe269fd2b70e8b20d7a90b7504ba`.

**Judgement calls** — dropped, though official or arguably so: staff of state bodies the list
does not name (directors of ministry departments, deputy presidents of NIK, agency heads), MPs of
the European Parliament, the presidential-office secretaries of state, a "Przewodniczący" without
a party tag, the Marshal's advisers, former MPs and ministers, KRRiT members, bare `Zastępca szefa
KPRM` and `Pełnomocnik rządu ...` labels without a state-secretary rank, the deputy chair of the
MON subcommittee on the reinvestigation of the air accident (`Pierwszy zastępca przewodniczącego Podkomisji ...`,
not an MP), secretaries of works councils, of the Komisja Wspólna Rządu i Samorządu Terytorialnego
and of the Rada Młodzieżowa, and the bare `Sekretarz stanu NAME` with no body named. Allowlist gaps
closed after the cut exposed them (maintainer's standing approval for false exclusions of MPs, government members
and Sejm/Senate staff): `Legislatorka`, `Sekretarz Komisji NAME`, `Senator RP`, `Ministra`, `Szef
KPRM`, `Członek Rady Ministrów`, `Wicemarszałkini`, BEOS/BAS specialists and experts with a topic,
deputy chief of the Chancellery, `P.o. dyrektora`, state secretaries holding a government office,
double-hyphen names and eleven exact labels (typos, name-plus-party forms; fourteen after the second
round). Second round (91 pieces,
158,442 characters): the chair of the MON subcommittee on the reinvestigation of the air accident, an MP, in two
exact labels without "poseł" and one with it (21 pieces, 66,583 characters); state secretaries with
the rank in any position (`Pełnomocnik rządu ..., sekretarz stanu w MP NAME`, `Zastępca szefa KPRM
podsekretarz stanu NAME`, `NAME podsekretarz stanu w ...`, one exact typo label `PodPodsekretarz`; 49 pieces, 89,234
characters);
`Sekretarz Komisji <Sejm committee> NAME` and `Starszy sekretarz Komisji` (18 pieces, 839
characters; the committee is matched against a list of Sejm committee names); `Dyrektor Biura
Obsługi Posłów` (3 pieces, 1,786 characters). Plain "Legislator NAME" is
kept: 99.7% of its characters belong to names that carry an explicit Biuro Legislacyjne label
elsewhere.

**Precision of the cut** (seeded sample, seed 20261008): 150 of 150 detected label lines were labels;
of 150 colon-terminated lines the detector rejected, 149 were prose and one a label with prose in
front of it on the same line (`... odwołuje Poseł Paweł Jabłoński (PiS):`), which stays inside the
previous piece. Known classes: prose ending in capitalised words before a colon is cut (the piece
is dropped: conservative); a label with an internal colon or wrapped over three lines is cut late or
not at all; a prose-prefixed label on one line is missed (an official's label is harmless, a guest's
would leave the words in the previous piece).

**Not established** — whether guest statements fall under art. 4 pkt 2 (a question for a
lawyer; the allowlist is the cautious option); that every kept label is an official acting in
office (the name slot accepts any two or three capitalised tokens); a rebuild from the PDFs
(no `pdftotext` or cache here: `build()` and `parse_pdf()` are tested on a synthetic cache and by
sharing the detector with the clean script, not on the PDFs); the NERGAL pass; about 845 footer-shaped
lines of real text (`2020 r.`) and 82 records quoting the transcript title in running text remain.

**Lesson** — a heuristic share of non-official text says how much to expect, not what to drop; and
a speaker filter is only as good as the turn boundaries under it: the first allowlist filtered
labels correctly but kept a guest's words inside an official's turn because the splitter knew only
role-anchored labels. Cut at everything that looks like a label, then classify.

## sejm_api_2011_2022 — dedup re-pinned to the PPC TEI `parliamentary` (2026-10-08)

**Finding** (slayerlabs/polish-dynaword-v2#54, option 5A) — the cross-source dedup scored
against the v0.2.5 `parliamentary`, which Hub PR 108 replaces with a rebuild from the PPC TEI
dump under a speaker allowlist. The rebuild keeps Sejm plenary text up to term 8 but drops
speakers the allowlist does not match, so a dedup against it keeps rows the old shard dropped.

**Fix** — `REFERENCES` pins the rebuilt `parliamentary.parquet` (`5f21252e…`). Reference groups
are read from the record id (`PPC_ID`, `PPC_KINDS`: house and directory code; plenary,
interpellation and reply codes are listed, every other code is a committee) instead of the old
heuristic. The duplicate outcome key `sejm_api_only` (same count as `only_sejm_api`) is removed.
126,967 rows enter the search as before; 92,369 are contained (93,861), 34,598 published
(33,106), 28,712,009 tokens. Against the old build 22 rows are newly dropped (2020-2022 Sejm
speeches 0.50-0.71 contained in a 2019-2023 Senate document) and 1,514 rows (7,126,113
characters) are kept again: 573 (4,908,128) whose speaker the allowlist drops (citizens'
initiative representatives, the President's Chancellery, the Prosecutor General's office, agency
heads; slayerlabs/polish-dynaword-v2#99), 470 (1,084,769) not in the dump (435 of them
"(tekst niewygłoszony)"), 357 (911,196) mostly in kept text but under 0.5 (315 lose mainly
floor interjections), 114 (222,020) to other rebuild drops. The 23 speeches of MP Piotr van der
Coghen stay dropped: the rebuild's name check keeps a member's name with a particle (decision
1A of 2026-10-08). Parquet sha256 `f6ec51dfd87b47b93905adcdd8f8dfb3f8449d6319df4153526c900bf258062f`.

**Re-pin (2026-10-08)** — Hub PR 111 repaired Windows-1250 misdecoding in `parliamentary` and
dropped 18,612 exact copies, so its parquet is `07903953…`. `REFERENCES` is re-pinned to it and
the script re-run on the restored inputs: the parquet, attribution and decisions are
byte-identical; only the pin in the stats changes. An exact copy cannot raise or lower the best
containment of a speech in one document, and the encoding repair changed no score in the
decisions (every speech scoring 0.3 or more, byte-identical).

**Not established** — the hand checks in the card still date from Hub PR 76. The 573 rows whose
speaker no allowlist rule matches stay here until the allowlist is applied to this source
(slayerlabs/polish-dynaword-v2#99).