Skip to content
harness.talk
Search

Sources & translations

Partial

What is served, what is declared

A source is either ingested with verification gates, declared with a spec for how it will be ingested, or planned. Nothing is paraphrased into a gap. The registry below is the same data /api/v1/sources returns.

Source registry

  • King James Bible, 1611 first edition

    kjv-1611
    kind
    text
    language
    en
    edition
    The Holy Bible, Conteyning the Old Testament, and the New — Robert Barker, London, 1611 (the 'He' Bible), original spelling
    licence
    Public domain. Body transcription and all witnesses are public-domain texts; no publisher's edition, notes or typesetting are reproduced.
    url
    github.com/aruljohn/Bible-kjv-1611
    • Body text of the reader and default for every API response. Spelling, punctuation, I/J and U/V usage, þ (thorn) and macron abbreviations are as printed in 1611, e.g. Gen 1:1 'In the beginning God created the Heauen, and the Earth.'
    • One transcriber-caused defect in the source was repaired with witnesses, never by hand: the transcription typed the divine name as 'Lord' where the folio prints it in capitals. LORD/GOD casing was restored by majority vote of three independent public-domain KJV witnesses — the CrossWire KJV module (casing derived from the Hebrew Strong's number behind each word: YHWH → LORD, Adonai → Lord), the eBible.org OSIS KJV and Project Gutenberg #10. A verse changed only when two witnesses agreed and their divine-name tokens matched the 1611 verse token-for-token. 104 verses with no such agreement (mostly 'my Lord' addressed to a man) were left exactly as transcribed and are listed in data/kjv/BUILD-REPORT.md.
    • Psalm superscriptions are stored as chapter titles (taken from the eBible OSIS witness; the transcription folded them into verse 1) and Psalm 119's Hebrew letter headings as headings. The translators' italics are not marked in the source.
    • The 1611 printing included the Apocrypha between the Testaments; it is not ingested. 66 books, 1,189 chapters, 31,102 verses, all gated in scripts/build-kjv.mjs; every input is SHA-256 locked in data/raw/SOURCES.lock.json.
    Ingested · verified

  • Geneva Bible, 1599 edition — verse text

    geneva-1599
    kind
    text
    language
    en
    edition
    The Bible, that is, the Holy Scriptures conteined in the Olde and Newe Testament — Geneva Bible, 1599 printing, original spelling
    licence
    Public domain (eBible.org distribution)
    url
    ebible.org/Scriptures/details.php?id=enggnv
    • The Reformers' English Bible, translated in Geneva 1557–1560 by Whittingham, Gilby, Sampson and others in Calvin's and Knox's circle; the Bible of Shakespeare, the Pilgrims and the 1611 translators' own study. Served beside the 1611 as a second text via ?source=geneva-1599 and the reader's Geneva line.
    • Verse text only. This source carries NO marginal notes, chapter arguments or cross-references — the reformers' commentary is a separate source (geneva-notes) and is not yet ingested. Nothing is substituted in its place.
    • The 1599 edition is served because it is the edition with a clean public-domain transcription; the 1560 first edition's verse text is nearly identical (the 1599 revised the notes, not the text). Psalm superscriptions are part of verse 1, as the Geneva printed them.
    • Versification differs from the KJV in 66 places (e.g. Job 41 is split differently, Numbers 12:16 falls in 13:1); the API returns not_found for KJV references with no Geneva counterpart rather than remapping. See data/geneva/BUILD-REPORT.md.
    Ingested · verified

  • Geneva Bible, 1560 first edition — facsimile

    geneva-1560
    kind
    facsimile
    language
    en
    edition
    The Bible and Holy Scriptures conteyned in the Olde and Newe Testament — printed at Geneva by Rowland Hall, 1560 (the first edition)
    licence
    CC0 1.0 / public domain (Internet Archive item '1560-the-geneva-bible')
    url
    archive.org/details/1560-the-geneva-bible
    • The true first edition: real page photographs of the 1560 Geneva Bible — the first English study Bible, the first with verse numbers, printed in readable roman type, with the reformers' marginal notes in the margins of the actual page. Served as a facsimile page-image viewer (Internet Archive IIIF), leaf by leaf, 1,224 leaves.
    • This source carries page IMAGES, not transcribed text. The only text layer the archive item has is a 2012/2024 OCR of the 1560 type, and it is unreliable (long-s ſ read as f, the two columns and side notes fused together); it is NOT served as scripture or as note text. Verse text is served from the clean 1599 transcription (geneva-1599); a separate, clearly labelled experimental layer (geneva-notes) reconstructs the note text.
    • Chapter-to-leaf mapping is approximate and opening-level, not verse-level: it points at the leaf where a book or nearby chapter opens, calibrated against leaves inspected by hand (e.g. Genesis opens at leaf 7). Browse forward/back from there; every image links to the same leaf on archive.org.
    Partially ingested

  • Geneva Bible, 1560 first edition — verse text (hand-transcribed)

    geneva-1560-text
    kind
    text
    language
    en
    edition
    The Bible and Holy Scriptures conteyned in the Olde and Newe Testament — printed at Geneva by Rowland Hall, 1560 (the first edition), original spelling
    licence
    Public domain (text)
    url
    archive.org/details/1560-the-geneva-bible
    • The verse text of the true first edition, hand-transcribed directly from the 1560 page images (Internet Archive IIIF), original spelling preserved exactly as printed — long-s, u/v and i/j as set, the ampersand and the yͤ / yͭ contractions, e.g. Gen 1:1 'In the beginning God created ye heauen and the earth.' Served in the reader as a labelled 'Geneva 1560' column beside the 1611 and the 1599.
    • Not OCR. The archive item's own machine text of the 1560 roman type is unreliable (long-s ſ read as f, the two columns and side notes fused) and is never served. Every verse in this source was read off the page and is checkable against it; words illegible in the imaging are left out of the chapter, not reconstructed.
    • Transcription in progress, chapter by chapter. Only the chapters present in data/geneva-1560/ are served; for every other chapter the reader falls back to the 1560 facsimile page image (source geneva-1560), and the column says so plainly rather than substituting the 1599 text in its place.
    • The 1560 and 1599 verse text are nearly identical (the 1599 revised the notes, not the text); this column exists so the first edition can be read and compared in its own spelling, not as a substitute witness. Ingested and human-verified so far: Genesis 1-5.
    Partially ingested

  • Geneva Bible marginal notes and chapter arguments

    geneva-notes
    kind
    annotations
    language
    en
    edition
    Geneva Bible, 1560 first edition (Rowland Hall, Geneva) — marginal notes, chapter arguments and cross-references
    licence
    Public domain (text)
    url
    archive.org/details/1560-the-geneva-bible
    • The reformers' own commentary in the margins: the chapter arguments, the doctrinal and explanatory notes, and the cross-references. This is the layer that made the Geneva Bible the people's study Bible and the one King James disliked.
    • Transcription in progress, book by book. Ingested and human-verified so far: Jonah (book argument, four chapter arguments, and the marginal notes legible in the facsimile). These were hand-transcribed from the 1560 first-edition page images (Internet Archive IIIF, leaves 748–750), original spelling preserved, and are served as machine-readable text at /api/v1/annotations/{ref}?source=geneva-notes and in the reader's margin. Every book not in data/annotations/geneva-notes/manifest.json still returns source_not_ingested — nothing is substituted.
    • Not a facsimile OCR. The archive item's own OCR of the 1560 blackletter is too mangled to serve (long-s ſ read as f, columns and side notes fused); it was never ingested. The transcription is done by reading the page images directly and is checkable against them. Notes that were clipped or illegible in the source imaging are omitted, not reconstructed.
    • For every book, and for the notes not yet transcribed, the marginalia come back as printed through the facsimile page-image viewer (source geneva-1560): each chapter in the reader opens the real first-edition leaf with the notes in place. Completing the transcription for the rest of the canon is ongoing work.
    Partially ingested

  • Textus Receptus (Greek NT)

    textus-receptus
    kind
    original-language
    language
    grc
    licence
    Public domain
    • The Greek textual tradition underlying the KJV New Testament (Erasmus 1516 → Stephanus 1550 → Beza 1598 → the 1611 translators).
    • Not yet ingested. The Greek text currently served (sblgnt-morphgnt) is a critical text and differs from the TR at known places (e.g. 1 John 5:7).
    • Ingestion spec: Scrivener 1894 or Stephanus 1550 plain-text with verse ids; verify against KJV verse boundaries.
    Declared · not ingested

  • Hebrew Old Testament — Westminster Leningrad Codex (OSHB morphology)

    wlc-oshb
    kind
    original-language
    language
    hbo
    licence
    CC BY 4.0 (Open Scriptures Hebrew Bible)
    url
    github.com/openscriptures/morphhb
    • 23,215 verses, word-level with Strong's number and OSHB morphology code; 200 verses tagged Aramaic (Daniel, Ezra, Jer 10:11, Gen 31:47) are kept in place with a per-verse language flag.
    • Ingested via the owner's Immutable/BibleOS data (scripts/build-originals.mjs); served at /api/v1/verses/{ref}?source=wlc-oshb and /api/v1/words/{ref}.
    • Verse boundaries follow the Hebrew numbering, which differs from the KJV in 100+ chapters (Psalm titles counted as verse 1, Joel/Malachi/1 Kings chapter splits). The API returns the Hebrew verse at the same chapter:verse number without remapping; the reader labels it. See data/originals/BUILD-REPORT.md for every difference.
    Ingested · verified

  • Greek New Testament — SBLGNT with MorphGNT tagging

    sblgnt-morphgnt
    kind
    original-language
    language
    grc
    licence
    SBLGNT: CC BY 4.0; MorphGNT: CC BY-SA 3.0
    url
    github.com/morphgnt/sblgnt
    • 7,928 verses, word-level with lemma, MorphGNT parsing code and part of speech. Ingested via the owner's Immutable/BibleOS data (scripts/build-originals.mjs).
    • A critical (eclectic) Greek text, NOT the Textus Receptus behind the KJV. Verified difference: 1 John 5:7 reads only '���τι τρεῖς εἰσιν οἱ μαρτυροῦντες,' — the Comma Johanneum is absent. 21 chapters have fewer verses than the KJV (e.g. Mk 9, Jn 8, Ac 8) because the KJV verses rest on TR readings the critical text omits.
    • Served for word study beside the KJV, always labeled 'Greek (SBLGNT)'. Requests for the TR return source_not_ingested until that text is ingested.
    • SBLGNT's inline apparatus sigla (⸀ ⸂ ⸃ … U+2E00–U+2E0D, pointers into its textual notes) are stripped from the served text; the apparatus itself is not ingested.
    Ingested · verified

Textus Receptus lineage

The 1611 translators worked the New Testament from the printed Greek of Erasmus (1516–1535), Stephanus (1546–1551) and Beza (1565–1604) — the line later called the Textus Receptus. The Old Testament came from the Hebrew Masoretic text of the Bomberg rabbinic Bibles. harness.talk names this lineage because it is the historical basis of the text we serve; it does not claim to serve the Greek itself until a verified TR edition is ingested (registry entry textus-receptus).

Transliteration framework

For multi-language agents the useful unit is not a translation but a stable address plus an exact source text. The framework is therefore: (1) every verse addressed by quartet and canonical ref, (2) each source text stored verbatim with its own id, (3) alignment tables between sources keyed by ref, and (4) transliteration performed by the agent, not by us, with the source id carried in the output. Alignment tables ship with the originals ingestion.

Specification · not yet live

Semantic mapping

Cross-source mapping (KJV ↔ Geneva ↔ Hebrew/Greek) is keyed by ref and, where morphology is available, by word index. It is data, not inference: an agent reading /api/v1/words/jn.3.16 will get the aligned Greek tokens with lemma and parsing, never a machine gloss presented as text.

Specification · not yet live

Comparison

Side-by-side comparison arrives with the second ingested text source. The reader is already structured for it — the marginalia column takes Geneva notes now and a parallel text toggle later. Until then, the honest state is one text, exactly.