1000_novels: rescreen with cited translator death years, pin the screen (837 -> 887 texts)

#65
by ppuzio - opened
Fabryka AI org
•
edited 3 days ago

What and why

Rebuilds 1000_novels from the v0.2.5 Parquet with an updated per-text rights screen, for slayerlabs/polish-dynaword-v2#44. PR #49 shipped the first screen. That screen had a defect: a known public-domain translator cleared an author whose death year was unknown. It also left 146 texts unresolved for want of translator death years.

Follow-up (maintainer decision): a Wolne Lektury public-domain footer no longer keeps a row. 1000_novels_0734 (translator "A.L.", not identified) is now dropped as unresolved, and the screen CSV is pinned by SHA-256 in the clean script.

Depends on an unmerged polish-dynaword-v2 change. The screen CSV evidence/1000_novels_rights_screen.csv has sha256 7fefc75f8fcbd07e5ec98842276765ffef9520b6acb4e24b4751c67cc3d2081c. It was written by scripts/novels_rights_screen.py on v2 branch v026-1000-novels-screen (off v2 origin/main), which is not yet committed. src/clean_1000_novels.py pins it as SCREEN_SHA256 and exits on a mismatch. license-evidence.json and stats.json record the same value, and a contract test checks that they match. Merge this PR only once that v2 change lands with this exact CSV.

Before → after

before (main, #49) after
documents 837 887
characters 129,722,658 130,201,301
tokens (cl100k proxy) 51,765,528 51,981,548

The text, added, created and token_count are unchanged on the 835 rows in both releases (checked row by row). Only 3 author credits change: 0144 and 0151 (A. W → A. W. (Antoni Lange)) and 0733 (Roux (unidentified) → Saint-Pol-Roux). 52 rows are added. Two rows are removed: 1000_novels_0462 (in copyright) and 1000_novels_0734 (unresolved).

screen status before after
pd_by_term (kept) 825 887
pd_per_wolne_lektury (kept) 12 0 (status removed)
in_copyright (dropped) 17 18
unresolved (dropped) 146 95

Rules, counts and hand checks

  1. Unknown author fails closed unless the work is listed as anonymous (anonim1–anonim5). Against main's CSV this changes 2 rows: 0216 (already dropped; only its basis changes) and 0733 (it would have been dropped). Hand check: I diffed the CSV before and after this rule alone, and only these 2 rows differ.
  2. Translator death years from authority files. There are 11 new entries, each citing a Biblioteka Narodowa (BN) authority id, a VIAF id and a BN catalogue record. The citations are inline in the screen script; the table is below. Hand check: I read every one of the 64 rows that changed against the rule-1 CSV, and none is a false name match.
  3. Cryptonym "A. W." resolves to Antoni Lange (d. 1929), for Chekhov only: 29 rows. BN bib 2019166 (Śmierć urzędnika, Bibljoteka Groszowa 1926) reads "krypt. tł.: A. W., nazwa właśc. tł.: Antoni Lange". BN authority 1220170 lists "A. W. (1862-1929)" under Lange. Hand check: I read the Wolne Lektury source line for all 29 texts, and each one names that 1926 edition. The same initials under any other author still fail closed (tested).
  4. Self-translation: 1 row. 1000_novels_0981 is Seneca's letter headed "PRZEKŁAD" with no name. It is Żuławski's own translation, item 9 of his collection Bajka o człowieku szczęśliwym (Orgelbrand 1914), checked against pl.wikisource and the Commons scans. It is now credited Lucius Annaeus Seneca (tłum. Jerzy Żuławski).
  5. A credited translator who died after 1955 makes the text in copyright: 1 row. 1000_novels_0462 (Wola mocy) is credited "tłum. Konrad Drzewiecki, Stefan Frycz". BN bib 1826277 lists both as translators, and Frycz died in 1960 (BN 1210480, VIAF 57800683). The row is dropped even though its Wolne Lektury footer says public domain. The screen's footer cross-check allowlists this one file by name and asserts that the allowlist entry is not stale.
  6. "Roux" identified as Saint-Pol-Roux (d. 1940) (BN 2977607, VIAF 24596597). 0733 is therefore kept on a cited death year.
  7. Hub script: src/clean_1000_novels.py gains 2 titles read by hand. 0697 has a title broken over lines. 0981 has a misspelt file name (senki). Titles: file 838, position 32, manual 17.
  8. No row is kept on a Wolne Lektury footer alone (maintainer decision): 1 row. The screen's pd_per_wolne_lektury path is deleted, not special-cased, so an unidentified credited translator fails closed like any other name without a death year. 1000_novels_0734 (Schopenhauer, Psychologia miłości, "tłum. A.L.") moves to unresolved. Hand check: the regenerated CSV differs from the previous one in this row only. The clean script's KEEP is now {"pd_by_term"}.

Translator outcomes

translator rows death year source
Józef Pracki 1 (0294) 1882 BN 1061302, https://viaf.org/viaf/162606612, iPSB
Edmund Cięglewicz 1 (0450) 1928 BN 1085439, https://viaf.org/viaf/101910807
Stanisław Sierosławski 2 (0211, 0292) 1941 BN 1049818, https://viaf.org/viaf/101848228
Jerzy Żuławski (self) 1 (0981) 1915 https://pl.wikisource.org/wiki/List_Seneki_do_Lucyljusza
A. W. = Antoni Lange 29 1929 BN bib 2019166, BN 1220170, https://viaf.org/viaf/68937797
Bolesław Londyński 22 1928 BN 1154157, https://viaf.org/viaf/101734529, PSB t. 17
Ludwik Jenike 1 (0208) 1903 BN 1438640, https://viaf.org/viaf/88990112
Adam Stögbauer 2 (0318, 0735) 1916 BN 1153547, https://viaf.org/viaf/20355997
Kazimierz Bronikowski 1 (0418) 1909 BN 1058992, https://viaf.org/viaf/164026788
Maria Cumft-Pieńkowska 1 (0464) 1920 BN 1683706, https://viaf.org/viaf/162253749
Izydor Berman 1 (0730) 1942 BN 1116279, https://viaf.org/viaf/294305466
Konrad Drzewiecki + Stefan Frycz 1 (0462) 1922 + 1960 BN 2607449 / 1210480, bib 1826277
"A.L." 1 (0734) unverifiable BN bib 1857221 gives the cryptonym only; dropped as unresolved

BN ids resolve at https://data.bn.org.pl/api/institutions/authorities.json?id=<id> (authority) and .../bibs.json?id=<id> (bib).

Provenance and licence

  • Upstream: CLARIN-PL handle 11321/312, pulled via SpeakLeash's redistribution. CLARIN-PL says the record holds metadata only, so no licence covers the texts (unchanged since #55).
  • Every kept row is public-domain: public domain in Poland under Art. 36 (life + 70).
  • No kept row rests on a Wolne Lektury statement alone.
  • Edition rights (Art. 99¹–99²) are not assessed.
  • The wolne_lektury / eltec_pol dedup is out of scope (#45).
  • The datasheet, license-evidence.json (release_scope), stats.json, the attribution sidecar, the sources.py entry (the legal_note sentence on rows relying on the Wolne Lektury footer is removed, and traceable no longer names the footer as a basis) and a findings-log entry are updated. release is unchanged.

Tests

  • Hub, before (92b36dc): 22 failed, 234 passed, 3 errors.
  • Hub, after: 21 failed, 235 passed, 3 errors.
  • The only difference is that src/test_1000_novels_contract.py::test_schema_counts_and_attribution now passes. It had failed on the attribution sidecar's LFS pointer.
  • The remaining failures are the same set before and after: LFS-pointer parquets of other sources in this worktree.
  • Rebased onto main c2099e6 (#61–#63 landed meanwhile; only the source_findings.md append conflicted, resolved by keeping both sections): 21 failed, 246 passed, 3 errors, same failure set.
  • Follow-up commit ead9e97: 21 failed, 248 passed, 3 errors, same failure set. The 2 new passes are the pin tests: SCREEN_SHA256 equals license-evidence.json and stats.json, and build() rejects an unpinned CSV before writing anything.
  • Merged main a8792dc (#64, #66; --no-ff; source_findings.md conflict resolved by keeping every entry, main's first) as d705fa1: 22 failed, 264 passed, 3 errors. The one new failure is src/test_global_voices_contract.py::test_release_has_no_partner_republications, which came in from main. It fails because its parquet is an LFS pointer in this worktree, like the others.
  • v2 scripts/test_novels_rights_screen.py: 10 passed. 6 fail against the v2 origin/main script. The new test_wolne_lektury_footer_does_not_rescue_an_unidentified_translator fails when the footer path is put back.

Sample (python3 src/render_sample.py data/1000_novels/1000_novels.parquet -n 3)

id: 1000_novels_0861
source: 1000_novels
added: 2026-06-26
created: 1785-01-01, 2013-12-31
token_count: 6189
license: public-domain
author: Rabindranath Tagore (tłum. Franciszek Mirandola)

Rabindranath Tagore

Głodne kamienie
Powrót
tłum. Franciszek Mirandola



Fatik Czakrawarti był hersztem szajki wiejskich bandytów. Nowe szelmostwo wpadło mu do głowy. W nadrzecznym mule leżał gruby, ciężki pień drzewa, czekając, aż go cieśle przemienią w maszt okrętowy. Fatik rozkazał swym oberwańcom wytężyć wszystkie siły, by pień ruszyć z miejsca i przetoczyć go gdzie indziej. Zamierzał w ten sposób zdziwić i doprowadzić do pasji właściciela pnia, a potem śmiać się z niego do rozpuku. Wszyscy poparli tę propozycję i zgodnie zabrali się do dzieła.

Ale właśnie w chwili gdy figiel miał zostać wprowadzony w czyn, zjawił się młodszy brat Fatika, Makhan. Nadszedł powoli, leniwym krokiem i bez słowa usiadł na pniu.

Chłopcy w pierwszej chwili oniemieli. Jeden z szajki trącił […]
id: 1000_novels_0503
source: 1000_novels
added: 2026-06-26
created: 1785-01-01, 2013-12-31
token_count: 183109
license: public-domain
author: Eliza Orzeszkowa

ELIZA ORZESZKOWA



Argonauci-strona tytułowa-separator 1.jpeg




ARGONAUCI





POWIEŚĆ




TOM I





Argonauci-strona tytułowa-separator 2.jpeg





WARSZAWA

NAKŁAD GEBETHNERA I WOLFFA
KRAKÓW – G. GEBETHNER I SPÓŁKA

1900





KRAKÓW. — DRUK W. L. ANCZYCA I SPÓŁKI.


I.
 Było to mieszkanie milionera. Na sprzętach i ścianach salonów barwy i połyski przelewały się jak na powierzchni konchy perłowej. Zwierciadła odbijały w sobie obrazy, posadzki lśniły jak zwierciadła. Tu i owdzie kobierce ciemne i firanki ciężkie przytłumiały świetność, ale na pozór tylko, w rzeczy samej podnosząc ją do uroczystości prawie kościelnej. Gdzieindziej wszystko jaśniało, błyszczało, mieniło się błękitem, szkarłatem, złotem, bronzem, rozmaitymi odcieniami białości, które wydają z siebie: gips, marmur, mora, […]
id: 1000_novels_0048
source: 1000_novels
added: 2026-06-26
created: 1785-01-01, 2013-12-31
token_count: 20929
license: public-domain
author: Michał Bałucki

Michał Bałucki
 


 

Grześ niepiśmienny
 


 


 


 


 


 

Rozdział pierwszy. Kto był Grześ i co umiał
 


 

Ojciec Grzesia nie był od tego, żeby jego chłopak umiał czytać i pisać i miał najszczersze chęci posyłać go do szkoły; ale zawsze się tak nieszczęśliwie składało, że nigdy na to czasu nie było. Zimy jedna po drugiej były takie ciężkie, mroźne, z zaspami śnieżnymi i zadymkami, że Wojciechowa — matka Grzesia, nie miała serca na taki czas gnać chłopca o pół mili blisko na pociejowskie grunta do szkoły; a przyszła wiosna, to zawdy chłopak był potrzebny w domu: to gęsi przypilnować na pastwisku, żeby w szkodę nie poszły, to po wódkę do karczmy, to małą Petrysię pokołysać, albo starej Giertrudzie pomagać przy bydle. W żniwa jeszcze bardziej był przydatny: gdy wszyscy wyszli w pole […]

🤖 Generated with Claude Code

ppuzio changed pull request status to open
ppuzio changed pull request title from 1000_novels: rescreen with cited translator death years (837 -> 888 texts) to 1000_novels: rescreen with cited translator death years, pin the screen (837 -> 887 texts)
ppuzio changed pull request status to merged
ppuzio deleted the refs/pr/65 ref

Sign up or log in to comment