01Why it exists
Thirty translations, four source languages
Hebrew, Greek, Aramaic and Latin sit alongside their English descendants, so agreement can be measured across the whole transmission chain rather than within one branch of it. Omnilogos turns that into a lattice you can walk: book → chapter → verse → word, every node linked, and every verse scored by how consistently independent lines render it.
02Parts list
- Fetch and ingestComplete JSON per translation from the scrollmapper/bible_databases set, no auth. Cached files don't re-download and ingested translations don't re-import, so you can interrupt it anywhere.
- Normalise and tagSpelling, archaic forms and divine names are folded so loued, loveth and love meet. Twenty themes shipped with 178,666 tags; the stage-8 pass widened it to thirty and reached 253,164 before the run was interrupted.
- CorrelateEvery pair of translations is measured verse by verse. The shipped model uses Jaccard similarity; stage 8 moves to TF-IDF cosine, because Jaccard lets short verses score high on shared function words.
- Model and exportA Bayesian score over cross-translation agreement (the corpus mean is 42.5%), exported as 200+ wikilinked markdown books, so Obsidian's graph view renders the whole corpus as a structure, not a search box.
03Construction
The sample on this page
Twenty-four verses in fifteen texts: the Westminster Leningrad Codex, the Textus Receptus, the Clementine Vulgate and twelve English lines from Tyndale (1525) to the Bible in Basic English, all public domain and taken verbatim from the same source the pipeline ingests.
Normalising, TF-IDF, pairwise cosine and the attestation score all run live in your browser. Pairs inside one transmission line count for less, because Geneva agreeing with Tyndale, which it revised, is weak evidence.
Lessons learned the hard way
The first source, getbible.net, died on an expired SSL certificate partway through. Long sleep chains over the SMB mount blocked everything, and logs written to the share made it worse. Now it launches with nohup, logs to /tmp, and gets checked on separately.
Not built yet
sentence-transformers embeddings for meaning-level agreement, a Greek interlinear, and a search.py CLI. The force graph on this page is the first pass at the graph view that was on the list.
04Tolerances
Attestation measures how consistently independent transmission lines render a verse, nothing more, and the vault says so in plain language. Lexical agreement is a proxy: two translations can use different words for the same sense (BBE's controlled vocabulary scores low for exactly that reason), which is what embeddings are for. Psalm numbering differs between traditions, and the sample is aligned by content. The page carries 24 verses; the full corpus lives locally.
what it is not, said plainly
05Inscription
Every node is a markdown file with wikilinks. The concordance is a place you can walk.
06The same room
- BallastEvery bill on a real calendar, ninety days projected forward, and one honest number: what's safe to spend today
- Sleep CompressionA sleep-restriction protocol turned tracker
- CryptoVaultEncryption you can read to the bottom of: RSA-4096 and AES-256-GCM in one HTML file, with identities, contacts, signatures and files
- Worm EaterAn ingredient scanner for food, cosmetics and personal care that says "I don't know" instead of quietly scoring the unknowns as safe
- DocCrawlerA local document library for the phone
- Mechanic's ArsenalMachines you can take apart in the browser