verify_paper_claims: read the v0.2.6 paper draft (v2#61)

- PAPER_NAMES: the 15 display names v0.2.6 adds.
- tab:additions: a minus typeset as $-$ is a negative cell, and a
"Net change" row is the totals row of a release that also removes sources.
- norm_license: a LicenseRef compares by its name, so "EU reuse policy"
matches LicenseRef-EU-Reuse-Policy (the eurlex FAIL).
- released_at: a source stops counting from its `removed` release. The three
withdrawn sources (wikisource, wikiquote, european_hplt_v3_pl) get
removed="0.2.6"; AGENTS.md and the source-rights-review skill say to set it.

Left for the release commit: v0.2.6 in src/plot_version_growth.py,
`release` on the 15 new sources (after NERGAL), the CHANGELOG section.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

.agents/skills/source-rights-review/SKILL.md

@@ -207,7 +207,8 @@ As done for HPLT (Hub commits 23e883a, 1802420, bade8e7):
 
 1. Remove `data/<key>/` data files from main.
 2. Keep the `SOURCES` entry for history; add a `legal_note` with the reason and the clause URL. Keep its old
-   `release` value (`european_hplt_v3_pl` still says `"0.2.3"`).
+   `release` value and set `removed` to the first release without it (`european_hplt_v3_pl`: `"0.2.3"` and
+   `"0.2.6"`). `verify_paper_claims.py` stops counting the source from `removed` on.
 3. Write a decision record (`status: excluded`, or `held_for_review` while a human decides).
 4. Run `python3 src/verify_paper_claims.py --tex ../polish-dynaword-v2/main.tex --ref worktree`. Its FAILs
    are the paper's token, document and source counts that the removal breaks.

AGENTS.md

@@ -208,7 +208,8 @@ Two rules about all of the above:
 ## Cut the release (maintainer, one commit after the source PRs merge)
 
 0. In `src/sources.py`, set `release` on each merged source this release ships
-   (check its shard went through NERGAL first). A source is not released until
+   (check its shard went through NERGAL first), and `removed` on each source it
+   drops. A source is not released until
    its rights review is recorded (see "Rights review").
 1. In `src/make_docs.py`: bump `VERSION` and `RELEASE_DATE`. Add a row for the new
    version to the version table and a row for each new contributor to the

src/sources.py

@@ -158,6 +158,7 @@ SOURCES = {
     "wikisource": {
         "added": "2026-06-15",
         "release": "0.2.0",
+        "removed": "0.2.6",
         "speakleash_key": "plwikisource",
         "pretty": "Polish Wikisource",
         "license": "CC-BY-SA-3.0",
@@ -580,6 +581,7 @@ SOURCES = {
     "wikiquote": {
         "added": "2026-06-15",
         "release": "0.2.0",
+        "removed": "0.2.6",
         "file_key": "plwikiquote",
         "pretty": "Polish Wikiquote (quotations)",
         "license": "CC-BY-SA-3.0",
@@ -916,6 +918,7 @@ SOURCES = {
     "european_hplt_v3_pl": {
         "added": "2026-07-14",
         "release": "0.2.3",
+        "removed": "0.2.6",
         "file_key": "european_hplt_v3_pl",
         "pretty": "HPLT v3.0 Polish (web, top WDS bins 10-5)",
         "license": "CC0-1.0",

src/test_verify_paper_claims.py

@@ -2,6 +2,7 @@
 
 import json
 from pathlib import Path
+import shutil
 import subprocess
 
 import pytest
@@ -182,6 +183,53 @@ def test_unreleased_source_is_left_out(repo, capsys):
     assert any("totals over ALL sources" in line for line in lines(out, vpc.WARN))
 
 
+def remove_dziennik(root: Path) -> None:
+    """v0.1.2 removes dziennik_ustaw: the data goes, the registry marks it, the tables follow."""
+    shutil.rmtree(root / "data" / "dziennik_ustaw")
+    edit(root / "src" / "sources.py", '"release": "0.1.0"}', '"release": "0.1.0", "removed": "0.1.2"}')
+    edit(root / "src" / "plot_version_growth.py", '"v0.1.1"]', '"v0.1.1", "v0.1.2"]')
+    edit(root / "src" / "plot_version_growth.py", "300_000]", "300_000, 100_000]")
+    edit(root / "src" / "plot_version_growth.py", "3_000_000_000]", "3_000_000_000, 1_000_000_000]")
+    edit(root / "CHANGELOG.md", "# Changelog\n", "# Changelog\n\n## v0.1.2 (2026-01-03)\n\n- Removed `dziennik_ustaw`: 1 source.\n")
+    tex = root / "main.tex"
+    edit(tex, "v0.1.1 & 2 & 300,000 & 3.000\\\\\n", "v0.1.1 & 2 & 300,000 & 3.000\\\\\nv0.1.2 & 1 & 100,000 & 1.000\\\\\n")
+    edit(tex, " & Total added & 100,000 & 1,000,000,000\\\\\n",
+         " & Total added & 100,000 & 1,000,000,000\\\\\n"
+         "v0.1.2 & Dziennik Ustaw & $-$200,000 & $-$2,000,000,000\\\\\n"
+         " & Net change & $-$200,000 & $-$2,000,000,000\\\\\n")
+    edit(tex, "Dziennik Ustaw & 200,000 & 2,000.0 & 969 & legal & v0.1.0 & official records; Art.~4 basis & none & dropped\\\\\n", "")
+    edit(tex, "Total & 300,000 & 3,000.0 & & & & 2 sources", "Total & 100,000 & 1,000.0 & & & & 1 sources")
+
+
+def test_removal_release_passes(repo, capsys):
+    remove_dziennik(repo)
+    code, out = run(repo, capsys)
+    assert (code, lines(out, vpc.FAIL)) == (0, [])
+    assert any("] v0.1.2 Net change documents " in line for line in lines(out, vpc.OK))
+    assert any("] v0.1.2 source count " in line for line in lines(out, vpc.OK))
+    assert any("] v0.1.0 source count " in line for line in lines(out, vpc.OK))  # removal is not retroactive
+
+
+@pytest.mark.parametrize("mutate, label", [
+    (in_file("src/sources.py", ', "removed": "0.1.2"', ""), "v0.1.2 source count"),
+    (in_file("main.tex", "Net change & $-$200,000", "Net change & $-$199,000"), "v0.1.2 Net change documents"),
+    (in_file("main.tex", "Dziennik Ustaw & $-$200,000", "Dziennik Ustaw & 200,000"),
+     "v0.1.2 additions sum to the documents change"),
+], ids=["registry keeps the removed source", "Net change", "removal printed as an addition"])
+def test_wrong_removal_claim_fails(repo, capsys, mutate, label):
+    remove_dziennik(repo)
+    mutate(repo)
+    code, out = run(repo, capsys)
+    assert code == 1
+    assert any(f"] {label} " in line for line in lines(out, vpc.FAIL)), out
+
+
+def test_paper_names_are_registry_sources():
+    registry = vpc.load_registry(vpc.ROOT, "worktree")
+    assert sorted(set(vpc.PAPER_NAMES.values()) - set(registry)) == []
+    assert len(set(vpc.PAPER_NAMES.values())) == len(vpc.PAPER_NAMES)
+
+
 def test_dual_licence_warns(repo, capsys):
     edit(repo / "src" / "sources.py", '"license_spdx": "CC0-1.0"', '"license_spdx": "CC-BY-4.0"')
     code, out = run(repo, capsys)
@@ -246,6 +294,8 @@ def test_printed_precision_sets_the_tolerance(raw, actual, ok):
     ("CC BY 3.0", "CC-BY-4.0", False),
     ("CC BY", "CC-BY-SA-4.0", False),
     ("mixed CC BY/BY-SA/PD", "per-record upstream license", True),
+    ("EU reuse policy", "LicenseRef-EU-Reuse-Policy", True),
+    ("EU reuse policy", "LicenseRef-TED-Notice-Reuse", False),
 ])
 def test_licence_comparison(paper, registry, ok):
     assert vpc.compatible(vpc.norm_license(paper), vpc.norm_license(registry)) is ok

src/verify_paper_claims.py

@@ -8,7 +8,8 @@ a tagged release is checked against that release and not against later commits.
   1. Release trajectory (`tab:releases`, `tab:additions`): documents, tokens and
      source counts per release, and the additions that reconcile consecutive totals.
      Ground truth: the VERSIONS/DOCUMENTS/TOKENS literals in src/plot_version_growth.py,
-     the `release` field in src/sources.py, and the CHANGELOG.md source counts.
+     the `release` and `removed` fields in src/sources.py, and the CHANGELOG.md
+     source counts.
 
   2. Corpus composition (`tab:sources` and the prose): per-source documents, tokens
      and first release, the source set, and the totals of the latest release.
@@ -67,6 +68,22 @@ PAPER_NAMES = {
     "Global Voices": "global_voices",
     "NKJP1M": "nkjp1m",
     "Wiktionary examples": "wiktionary_examples",
+    # Added in v0.2.6.
+    "UZP rulings": "uzp_orzeczenia_pl",
+    "Wikipedia user talk": "plwiki_user_talk",
+    "Sejm committee transcripts": "sejm_committee_transcripts",
+    "Wikipedia talk": "plwiki_talk",
+    "TED notices": "ted_pl",
+    "Sejm speeches 2011--2022": "sejm_api_2011_2022",
+    "Sejm interpellations": "sejm_interpellations",
+    "Nonsensopedia": "nonsa",
+    "OPEN ICM": "open_icm_pl",
+    "Studia BAS": "studia_bas",
+    "Wiadomo\\'sci Statystyczne": "wiadomosci_statystyczne_pl",
+    "ROCK Politechnika Lubelska": "rock_pollub_pl",
+    "Edukacja Medialna": "edukacja_medialna_pl",
+    "Open AGH chemistry": "open_agh_chemistry_pl",
+    "Kultura bez Barier": "kultura_bez_barier_pl",
 }
 
 # Prose claims about the whole corpus: what they count, the printed unit, the pattern.
@@ -80,7 +97,9 @@ CLAIMS = (
 
 UNITS = {1: "", 1e6: "M", 1e9: "B"}
 
-NUMBER = re.compile(r"\d[\d,]*(?:\.\d+)?")
+TOTAL_ROWS = ("Total added", "Net change")
+
+NUMBER = re.compile(r"-?\d[\d,]*(?:\.\d+)?")
 VERSION = re.compile(r"v\d+\.\d+\.\d+")
 TABLE = re.compile(r"\\begin\{(table\*?)\}.*?\\end\{\1\}", re.S)
 
@@ -152,7 +171,8 @@ def _rows(tex: str, label: str, columns: tuple[str, ...]) -> list[dict[str, str]
     for line in _table_body(tex, label).splitlines():
         line = line.strip()
         if line.endswith("\\\\"):
-            cells = [cell.strip() for cell in re.split(r"(?<!\\)&", line[:-2])]
+            # A minus sign is typeset as $-$.
+            cells = [cell.strip().replace("$-$", "-") for cell in re.split(r"(?<!\\)&", line[:-2])]
             if len(cells) == len(columns):
                 rows.append(dict(zip(columns, cells)))
     return rows
@@ -165,9 +185,10 @@ def parse_releases(tex: str) -> dict[str, dict[str, str]]:
 
 
 def parse_additions(tex: str) -> tuple[dict[str, list[dict]], dict[str, dict]]:
-    """tab:additions: version -> its addition rows, and version -> its "Total added" row.
+    """tab:additions: version -> its addition rows, and version -> its totals row.
 
-    A row with an empty release cell continues the release above it.
+    A row with an empty release cell continues the release above it. The totals row
+    is "Total added", or "Net change" for a release that also removes or shrinks sources.
     """
     additions: dict[str, list[dict]] = {}
     totals: dict[str, dict] = {}
@@ -176,7 +197,7 @@ def parse_additions(tex: str) -> tuple[dict[str, list[dict]], dict[str, dict]]:
         if not (NUMBER.fullmatch(row["documents"]) and NUMBER.fullmatch(row["tokens"])):
             continue  # header
         release = row["release"] or release
-        if row["addition"] == "Total added":
+        if row["addition"] in TOTAL_ROWS:
             totals[release] = row
         else:
             additions.setdefault(release, []).append(row)
@@ -264,10 +285,11 @@ def load_registry(root: Path, ref: str) -> dict[str, dict]:
 
 
 def released_at(registry: dict[str, dict], version: str) -> set[str]:
-    """Sources whose `release` is at or before `version`."""
+    """Sources whose `release` is at or before `version` and whose `removed`, if any, is after it."""
     ceiling = _version_key(version)
     return {name for name, cfg in registry.items()
-            if cfg.get("release") and _version_key(cfg["release"]) <= ceiling}
+            if cfg.get("release") and _version_key(cfg["release"]) <= ceiling
+            and not (cfg.get("removed") and _version_key(cfg["removed"]) <= ceiling)}
 
 
 def _version_key(version: str) -> tuple[int, ...]:
@@ -350,7 +372,7 @@ def check_trajectory(report: Report, tex: str, root: Path, ref: str) -> None:
             added = sum(int(_number(row[what])[0]) for row in rows)
             report.num(f"{version} additions sum to the {what} change", added, change)
             if version in totals:
-                report.printed(f"{version} Total added {what}", totals[version][what], change)
+                report.printed(f"{version} {totals[version]['addition']} {what}", totals[version][what], change)
 
 
 def check_composition(report: Report, tex: str, root: Path, ref: str) -> None:
@@ -449,7 +471,8 @@ def norm_license(text: str) -> tuple[str, str | None]:
     match = re.search(r"CC[- ]BY(?:[- ](SA))?(?:[- ](\d+\.\d+))?", t)
     if match:
         return ("CC-BY-SA" if match.group(1) else "CC-BY", match.group(2))
-    return (t.strip(), None)
+    # A LicenseRef spells a licence's name: LicenseRef-EU-Reuse-Policy is "EU reuse policy".
+    return (re.sub(r"^LICENSEREF-", "", t.strip()).replace("-", " "), None)
 
 
 def compatible(a: tuple[str, str | None], b: tuple[str, str | None]) -> bool: