Sources
Provenance log for every source text on the site (build spec §0). The /sources page renders this file.
The Darby Translation (Tier 1, public domain)
Work. The Holy Scriptures: A New Translation from the Original Languages by J. N. Darby. New Testament 1867 (second edition 1872, third 1884); Old Testament 1890 (G. Morrish, London). Public domain in the United States.
Base text. ebible.org edition engDBY, USFM files dated 2019-11-15, downloaded 2026-09-21 (SHA-256 of the zip in docs/reports/bible-sources.sha256). ebible.org states the text is public domain. The same text is the CrossWire Sword module "Darby" (version 2.0, 2020-12-21), whose metadata says it derives from printed copies distributed by Bible Truth Publishers, Addison, Illinois. The edition adds Strong's numbers and 4,297 name-gloss notes (for example "Elohim" on "God"); neither is Darby's, and the build strips both. Darby's emphasized words are marked in the file as bold; the site renders them in italics, matching the printed editions (owner ruling 2026-09-22).
Integrity check. A second, independent lineage was compared verse by verse: the Unbound Bible (Biola, 2009) text as distributed in the open-bibles repository (eng-darby.zefania.xml, SHA-256 in the same report). STEM Publishing and CCEL carry the same lineage-B text (a shared typo in Job 5:14 proves common descent). Result on 2026-09-21: 66 books and 1,189 chapters in both; of 31,098 shared verses, 30,973 identical after normalizing emphasis and psalm-title placement; 116 differ only in psalm-title placement; 2 in spelling; 2 in spacing typos in lineage B; 5 substantive. The build re-runs this comparison and fails on any divergence beyond the logged set (research/bible/lineage-b-diff.json).
| Lineage | Editions | Provenance | Notes |
|---|---|---|---|
| A (base) | ebible.org USFM 2019; CrossWire Sword "Darby" | Bible Truth Publishers printed copies (Sword metadata) | Strong's and name-gloss notes added by the digitizers; British spelling; psalm titles as headings; emphasis marked |
| B (check) | Unbound Bible 2009; open-bibles Zefania; STEM Publishing; CCEL | Unbound Bible e-text | No notes; psalm titles inlined into verse 1; omitted verses as placeholders; STEM renumbers Acts 8 after the omitted verse |
Scan witness. The 1867 second-edition New Testament (G. Morrish), Princeton Theological Seminary copy, archive.org item DarbysWritingsNewTranslationNewTestamentSynopsisBooksBibleMulti. It confirms Acts 8:37 and 15:34 are footnote-only and reads "Nazaræan" at Matthew 2:23. No Old Testament scan has been reached: the archive.org item titled "1890" is a 2022 PDF generated from the ebible files, and HathiTrust sits behind a bot check.
Arbitrations. Recorded in research/bible/overrides.json with a basis of scan or sense. Matthew 2:23 corrected to "Nazaræan" (scan). Four Old Testament readings (2 Samuel 23:4, Job 5:14, Psalm 146:10, Jeremiah 45:2) keep the lineage-A reading on sense; their verses carry an internal "unverified-against-scan" flag until an Old Testament scan rules.
Versification. Follows Darby: Matthew 23:14, Acts 8:37 and Acts 15:34 are absent from the running text, each with a one-line note on the chapter page; 2 Kings 11 is numbered per lineage A.
Footnote markers. Nineteen verses of Psalm 119 in the e-text carry a bare asterisk before "word", the printed edition's footnote reference. The build removes the asterisk from the rendered text and records it in the verse's internal flags for the footnote layer.
Darby's Synopsis of the Books of the Bible (Tier 1, public domain)
Work. J. N. Darby, Synopsis of the Books of the Bible, five volumes (G. Morrish, London; first edition 1857–1867, revised editions following). Public domain in the United States.
Source. STEM Publishing, https://www.stempublishing.com/authors/darby/synopsis/, the source spec §0 names for Darby's writings. Every page was fetched once on 2026-09-23 at one request per second by tools/fetch-synopsis.py; research/synopsis/manifest.json records each page's URL, size and SHA-256. tools/build-synopsis.py keeps the text, headings, italics and bold, and removes STEM's navigation, links and scripts; nothing else is changed. Scripture references are linked to the reader at render time.
Mapping. Each STEM section is mapped to the chapter range its label names ("Chapter 1", "Chapters 6 to 8", "Chapters 9:8 to 12", "Psalms 9 and 10"). Chapter pages show the opening of every section that covers them and link to the full section at /bible/[book]/synopsis/[section]. STEM has no section of its own for seven chapters (Leviticus 12, Jeremiah 45–48, Matthew 20, Luke 23); in each case the neighbouring section's text treats the chapter, and the chapter page links there with a "treated within" note (owner ruling, 2026-09-23; evidence in docs/reports/synopsis-validation.md). Introductions, summaries, conclusions and the five Psalms book divisions are listed on the book page.
Deferred. Darby's own footnotes exist only in the printed editions and the scans. Plan item "Darby footnote layer — blocked on OT scan (owner)".
The library as text (Tier 1 and Tier 3b)
Scope. The public-domain expositions (Tier 1) and the assembly ministry (Tier 3b, by permission) read as HTML in the reading room; the Present Truth Publishers items (Tier 2) and the Tier 3 studies stay PDF-only, with the credit and link (owner ruling, 2026-09-23). Every HTML page links its PDF.
Extraction. The text is taken from each PDF's own text layer with pdfplumber (tools/extract-library.py): lines are rebuilt from word positions, running heads, footers and page numbers are removed, lines are rejoined into paragraphs, end-of-line hyphens are repaired, and italic and bold type are kept. Six files whose text layer is unusable (a Kelly volume and five Newberry lectures) are read by the macOS Vision recognizer from the page images; four Mackintosh pieces are scans whose text layer came with the PDF. Both kinds carry a visible note that the text was read from a scanned page. Nothing is edited: no wording, spelling or punctuation is changed, except a double hyphen shown as a dash.
Checks. tools/build-library.py compares every file's rendered words with the PDF text layer and fails the build below 98%; the report is docs/reports/library-validation.md.