Source Library First-Translation Verification Corpus (snapshot 2026-10-26)

The first-translation verification corpus of Source Library (sourcelibrary.org): documented evidence of absence at scale. 73,436 recorded verification attempts with per-attempt provenance (instrument, sources consulted, verbatim queries, result, evidence grade), graded verdicts on 10,369 books under an 8-verdict taxonomy, durable screening decisions, and the composition of the bibliographic reference set every catalogue-tier absence claim is asserted against. Each "first English translation" claim on the site derives from this ledger; the dataset makes every such claim independently auditable. See DATASHEET.md inside the deposit for motivation, composition, collection process, known biases (measured reference-set recall, instrument fabrication rates, the post-1950 blind spot), and recommended uses. Snapshot 2026-10-26, generated from git 081ffb30ddf656524c78037141e0729d658853d4 by scripts/eval/export-ft-dataset.mjs in the sourcelibrary-v2 repository (AGPL-3.0). Data license: CC BY 4.0. Contains no personal or reader data.