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.

---

## 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.