Dataset provenance

Every dataset Darash serves records which printed work it is, which edition, how the text reached us, under what licence — and what is not established about it.

Generated from the live list_dicts response on 2026-09-13. Nothing here is hand-written: call
list_dicts yourself and you get the same records, field for field.

How to read the identification field

One of those is a finding. The other two are inheritances. We label them differently because
they are different, and because a reader deciding how much weight a gloss can carry needs to
know which one they are holding.

A field reading UNKNOWN — not established is an honest gap. An empty field would be
uncheckable; a confident-looking guess would be worse than either.

The verified one: Thayer

A Greek-English Lexicon of the New Testament, being Grimm's Wilke's Clavis Novi Testamenti, translated, revised and enlarged
Joseph Henry Thayer (translator, reviser, editor); after C. L. W. Grimm's revision of C. G. Wilke's Clavis
Harper & Brothers, New York, 1889 — Corrected Edition. Read from the title page of our own scan, not assumed.

Method. Page-image transcription of all 716 body pages, two independent passes collated (98.91% agreement over 856,793 tokens), segmented into entries, every entry proven a verbatim, non-overlapping, in-order slice of its own page. References were modernised ("Matthew 27:24", never "Mt. xxvii. 24") under a PROSE INVARIANT: strip book names, abbreviations and numerals from before and after, and the remaining words must be identical. An entry whose prose changed by one word is REJECTED, not published, and 94 entries where no rewrite passed keep the deterministic expander's baseline instead. Each entry records which of the two produced its served text (modern_refs_source) and which rule produced its Strong's number (strongs_match: exact, normalised, second-pass, ambiguous or none). Imported into this blob by cmd/thayer_import, which verifies the source SHA-256 before it writes anything.

Source. git.dixt.io/nor/thayer-lexicon, build commit 24de0cb77ce73a677aa8b165f05d8478dfe2de65, file out/entries_4_final.jsonl (sha256 f03f0b8e5d2deda209247e597155c05d4599a0dcda30b14d9a67d3b1ab5a7376). Built from public-domain page scans of the 1889 print (archive.org), with a second physical copy of the same stereotype plates used to adjudicate contested glyphs.

What is and is not established. 5,870 entries; 5,529 carry a Strong's number (3,779 by exact headword match, 1,648 normalised, 102 by a second pass over unambiguous lemma variants). 341 do not, and 14 of those carry candidates rather than a guess — an ambiguous headword REFUSES to resolve. Verification is not uniform: every entry has been read by a model reader and the corpus is census-proven against the published dictionary (Greek characters and terminal asterisks exactly equal, sha bijection 5,870/5,870), but only a small number of pages have been collated line by line against the image by eye. Measured on the 13 pages that were independently transcribed twice, Greek-bearing token disagreement runs at 0.19% — roughly two per page — mostly accent, iota subscript and συν-/συγ- assimilation. One illegible Josephus digit is kept as "?": flagged, never filled. Reference modernisation is not uniform either: 5,776 entries are gemini-dual-witnessed, 94 carry the deterministic expander's baseline because no rewrite passed the prose invariant, and each entry says which it is. Citation triage found and fixed four real ink-fill digit errors (3 read as 8), each proven against the second copy's page image and never against Scripture; the residual ~660 citation flags are the measured benign classes — Thayer follows the critical text, darash indexes the Received Text.

A note on the asterisk. Thayer's terminal asterisk is CONTENT, not punctuation: it asserts that every New Testament occurrence of the word is cited in the entry. It is preserved, and the corpus build disqualifies any rewrite that changes an entry's asterisk count — a model once pattern-completed twelve of them onto entries whose pages have none. Thayer held Unitarian views; on deity-of-Christ passages (G2316, G166) read him as the careful philologist he was and take the theology elsewhere.

Per-entry ledger. out/verified.jsonl in git.dixt.io/nor/thayer-lexicon — per entry: the printed-text sha256, the pages, and what verified it. The same per-entry sha256 and pdf_pages travel on every DictEntry served here, so one definition can be checked against one page.

The Strong's numbers are our editorial addition, from STEPBible TBESG (CC BY 4.0). The 1889
print contains none.

Strong's lexicons

greek_lex — Strong's Greek Dictionary of the New Testament, augmented with STEPBible's Translators Brief lexicon (TBESG), the full Abbott-Smith Manual Greek Lexicon, and the Translators Formatted full LSJ (TFLSJ)

Source. number, lemma, transliteration, pronunciation, def_short, def_lit, derivation, see, comment and usage: bibledb MySQL table lexicon_greek (upstream distribution not established). def_long: STEPBible TBESG "Gloss" column. abbott_smith: full Abbott-Smith TEI XML (github.com/translatable-exegetical-tools/Abbott-Smith), falling back to TBESG "Meaning". lsj: STEPBible TFLSJ. step_gloss: TBESG "Gloss".

Method. SQL export via LoadFromMySQL → loadLexicon, then MergeSTEPBibleLexicons and MergeFullAbbottSmith. def_long was REPLACED on 2026-09-02 by cmd/lexicon_gloss_rebuild, a surgical rewrite of this one blob key that copies every other top-level field through as raw msgpack it never decodes — make sync was NOT run, because it silently drops any dataset whose sources are absent.

Known gaps. 5,523 entries. WHAT def_long USED TO BE IS THE GAP THAT WAS CLOSED: until 2026-09-02 it served the gloss set withdrawn from the thayer key the same day — probably Blue Letter Bible's Outline of Biblical Usage, public-domain status never established, arrived through bolls.life. It was archived verbatim first (sources/archived-darash-greek-lex-def-long/ in git.dixt.io/nor/thayer-lexicon, 5,523 entries, sha256 574dc4cb6b72da668a3fc90d375d86831426be918da1d2f327ef967c37d44566) and only then replaced. Still not established: which upstream distribution populated the MySQL table, when it was imported, and which STEPBible-Data commit the merges read. def_short, def_lit, derivation, comment and usage are Strong's own text and were not touched.

Note. A stale blob cannot bring the old def_long back: isWithdrawnGlossList fingerprints it at LOAD time by shape and the store empties the field rather than serving it.

hebrew_lex — Strong's Hebrew and Aramaic Dictionary of the Old Testament, augmented with STEPBible's Translators Brief lexicon (TBESH) and the OpenScriptures Brown-Driver-Briggs

Source. number, lemma, transliteration, pronunciation, def_short, def_lit, derivation, see, comment, usage and part_of_speech: bibledb MySQL table lexicon_hebrew (upstream distribution not established). def_long: STEPBible TBESH "Gloss" column. abbott_smith: OpenScriptures BDB article per homonym branch, with TBESH "Meaning" where longer. step_gloss: TBESH "Gloss".

Method. SQL export via LoadFromMySQL → loadLexicon, then MergeSTEPBibleLexicons and MergeFullBDB. abbott_smith was REGENERATED on 2026-09-02 by cmd/hebrew_bdb_rebuild through the FIXED import path (commit 08d6d35): LexicalIndex.xml's homonym letter is no longer discarded, so H2617 חֶסֶד now shows חֶסֶד I "goodness, kindness" AND חֶסֶד II "shame" instead of whichever cross-reference happened to precede the other in the file. 543 lemmas changed; every one of them gained a labelled homonym branch, and no single-sense lemma moved. def_long was REPLACED the same day by cmd/lexicon_gloss_rebuild. Both are surgical rewrites of this one blob key that copy every other top-level field through as raw msgpack they never decode — make sync was NOT run, because it silently drops any dataset whose sources are absent.

Known gaps. 8,674 entries. def_long carried the Hebrew half of the same withdrawn corpus as greek_lex. There is no archived Hebrew witness to diff against — darash never served it under a dictionary key — so the identification rests on three things, each recomputable by python3 scripts/measure_withdrawn_similarity.py in git.dixt.io/nor/thayer-lexicon, which states its method in full. (1) The import path: one loadLexicon reads the same data→def→long column out of both tables, and STEPBible enters through a different function into different fields. (2) The editorial signature both halves share — darash's copy expands abbreviations both upstreams leave short: 132 of the 8,674 entries contain "figuratively" and NOT ONE contains "fig.", against TBESH's 162 entries containing "fig." and none containing "figuratively". (3) What TBESH has that this text lacks: TBESH rewrote every proper name's gloss as Tyndale's own prosopography ("A man living at the time of Exile and Return, only mentioned at Est.9.8; son of: Haman…"), and 1,450 of the 8,131 comparable numbers carry it in TBESH while def_long still holds the old one-line gloss ("a son of Haman"). Text derived FROM TBESH cannot be missing 1,450 of TBESH's own articles. A SIMILARITY FIGURE CANNOT SETTLE THIS, and the one recorded here until 2026-09-02 ("NOT TBESH — 39.6% byte-identical, median similarity 0.806") pointed the wrong way as well as being unreproducible: remeasured against TBESH's "Meaning" column, stripped of numbering, HTML and cross-reference apparatus, the withdrawn text is 49.4% byte-identical with median similarity 0.986. Both descend from the same abridged Brown-Driver-Briggs — TBESH says so of itself — so distance was never going to separate them. Archived verbatim first (sources/archived-darash-hebrew-lex-def-long/ in git.dixt.io/nor/thayer-lexicon, 8,674 entries, sha256 df153764a3793c0b30424f331c667c8f5ca4972abcc3be288dbb46196cd8a88c). H3328 יִצְחַר carries no def_long and never did. Still not established: which upstream distribution populated the MySQL table, when it was imported, and which STEPBible-Data / HebrewLexicon commits the merges read. def_short and the other Strong's fields were not touched.

Note. Extended Strong's letters are the homonym axis and are preserved in both def_long and abbott_smith. A bare H-number can be two different words.

Dictionaries

13 dictionaries are served. 12 are described in full below.

wilsons is served but is not described here: it is the one source whose rights
position we have not established, and it is under review. We would rather say that than
publish a licence claim we cannot stand behind.

webster — Webster's Unabridged Dictionary of the English Language

Source. bibledb MySQL table entries (word, wordtype, definition), active rows only (upstream distribution not established)

Method. SQL export via LoadFromMySQL → loadWebster; wordtype prefixed to the definition as "(type) text"

Known gaps. EDITION IS THE GAP: darash's own tool text has called this "Webster 1828"; that is an assumption, not a finding. It is a general English dictionary, not a Bible dictionary.

isbe — The International Standard Bible Encyclopaedia

Source. bibledb MySQL table isbe (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

Known gaps. 1915 archaeology and higher-critical dating are dated; the 1979 revised ISBE is a different, copyrighted work and is NOT this

navestb — Nave's Topical Bible

Source. bibledb MySQL table navestb (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

Known gaps. verse references are KJV versification; not re-checked against darash's own book/verse grid

smith — Smith's Bible Dictionary

Source. bibledb MySQL table smith (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

Known gaps. which of the several Smith abridgments this is has NOT been established

easton — Easton's Illustrated Bible Dictionary

Source. bibledb MySQL table easton (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

Known gaps. upstream distribution and its transcription method not established; entry count not collated against a print edition

fausset — Fausset's Bible Dictionary

Source. bibledb MySQL table fausset (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

hitchcock — Hitchcock's New and Complete Analysis of the Holy Bible (Bible Names Dictionary)

Source. bibledb MySQL table hitchcock (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

Known gaps. name etymologies are 19th-century and frequently disagree with modern Semitic philology; treat as historical, not lexical evidence

ats — American Tract Society Bible Dictionary

Source. bibledb MySQL table ats (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

hawker — The Poor Man's Concordance and Dictionary to the Sacred Scriptures

Source. bibledb MySQL table hawker (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

Known gaps. devotional, not lexical; the author's own theology is present throughout and is not marked

torrey — The New Topical Text Book

Source. bibledb MySQL table torrey (upstream distribution not established)

Method. SQL export of (topic, text) rows via LoadFromMySQL → loadDictionary; no transcription, no editing

Known gaps. verse references are KJV versification; not re-checked against darash's own book/verse grid

bullinger — Figures of Speech Used in the Bible

Source. pre-extracted bullinger_clean.json, derived from a TheWord .dct.twm module (human-proofed text); see MergeBullinger for the search paths

Method. JSON import via MergeBullinger; entry Topic taken from the module's full_title field

Known gaps. the .twm module's own transcription accuracy has not been checked against the 1898 print edition

Corrections

2026-09-02 — a dictionary served under the wrong name

Until this date Darash served, under the key thayer, a text that was not Thayer. It was in all
probability Blue Letter Bible's Outline of Biblical Usage, whose public-domain status was
never established, and it reached us through a source later found to be unreliable. The same
withdrawn gloss set also populated def_long on both Strong's lexicons.

All three were archived verbatim, with digests, before anything was replaced:

thayer was then replaced with the transcription described at the top of this file. A stale
cache cannot bring the old text back: the withdrawn gloss set is fingerprinted by shape at load
time and the field is emptied rather than served.

Why this section exists

Because the provenance work is what found it. A feature list would not have. If you are
weighing whether to trust this data, the known_gaps field on every source above is written in
the same spirit — read it as the part we could not prove, not as boilerplate.

Generated from the live service's own tools/list — not written by hand, so it cannot drift from what Darash actually answers. Source on GitHub