Measured Scripture translation quality
Every verse carries the evidence of its accuracy.
AI can draft Scripture in a low-resource language. The unsolved problem is trust — and Logos Canon answers it per verse: an independent back-translation, two computed scores, severity flags quoting the exact words at issue, a Greek or Hebrew consultant check, and judges whose catch rates are themselves measured with planted errors.
Draft, check, review, certify — any language with a seed corpus.
Generation and judgment never share evidence
Every chapter passes through independent stages. The drafter sees helps the judges never see; the judges see originals the drafter is never trusted to have used correctly.
Retrieval-taught drafting, best of three
One verse at a time, taught by the most relevant verse pairs from the language's own published corpus — real orthography and style, not generic imitation. Three candidates are drafted and an independent check picks the winner.
Independent back-translation — the witness
A separate model renders the draft back into the gateway language in clinical literal mode, without ever seeing the source verse. An anti-recitation rule stops it from “remembering” the famous verse instead of reading the words in front of it — proven, not assumed.
Computed verdicts, never opinions
Deterministic safety nets run first. Then a judge emits severity-graded flags quoting the exact wording at issue — and the 0–100 score is computed from those flags by fixed weights. The model is never allowed to assert a number.
Original-language consultant
Every verse is audited against the Greek New Testament or Hebrew Old Testament directly, with word-level scholarly data — lemma, morphology, senses, referents, and textual-variant awareness, so manuscript-tradition differences are never flagged as errors. Flags distinguish draft errors from readings inherited from the gateway translation.
Second witness on serious flags
Any verse with a major flag gets a second, corpus-blind back-translation from deliberately different conditioning. Two witnesses that share a bias are one witness.
Human review, then certification
A mother-tongue reviewer approves, corrects, or rejects verse by verse, with per-verse plain-language guidance. Confirmed books can earn a bounded statistical claim — “at most X% of verses carry a meaning-changing error, at 95% confidence” — by the same acceptance-sampling method used in WHO field surveys.
Measured, not asserted
A judge you haven't tested is just another opinion. Every automated check in Logos Canon is fed planted known-answer errors — and known-clean verses — and the catch rates are printed on every report, re-runnable from the dashboard at any time. These are our latest production numbers.
41 of 43 planted subtle errors caught against the Greek — error classes derived from real translation hazards, validity-judged.
43 of 43 swapped-original probes flagged — the consultant demonstrably reads the verse in front of it instead of quoting memory.
Clean verses cry wolf rarely — measured on known-good text, with confidence intervals on every number.
Deleted clauses and spliced verses planted into healthy text: 30 of 32 caught by the adequacy gate.
The auto-repair stage was trapped trying to cheat during validation, hardened with guards, and only then shipped. After the guards: zero laundering — rejected repairs stay rejected.
Dropped “therefore”, inverted “but” — the logic between clauses is checked as strictly as the words, with zero new false alarms.
Audit an existing translation
The same battery runs in reverse: upload a published translation and every verse is scored against the original languages — nothing machine-filled, versification differences realigned so they are never scored as mistranslation. “How good is the Luke we already published?” is a question we have not seen any other tool answer at the verse level.
Verse-level verdicts
Two scores and quoted evidence per verse — not a single opaque grade for the whole book.
Inherited vs introduced
Where your text follows the gateway but the gateway departs from the Greek, the report says so — a tradition decision, never counted as your error.
Printable audit trail
Every report carries its methodology, data attribution, and the measured catch rates of the very checks that produced it.
Standing rules, enforced in code
Machine drafts are labeled MACHINE DRAFT — NOT SCRIPTURE on every report. No certification is issued before mother-tongue reviewers confirm the text, verse by verse. Nothing machine-made can enter an audit of a human translation. And the platform is language-agnostic by design: no language or script is hardcoded into any prompt or check — everything is parameterized per project or measured from the corpus itself.
The complete picture
Everything Logos Canon does today, on one page: what it is, how quality is built into every verse, the parts we have not seen any other tool do, and a walkthrough of a chapter from draft to report.
How quality is built in
Eight independent stages from retrieval-taught drafting to statistical certification, with the vocabulary pass that measures drafts against the team's own verses.
What we have not seen elsewhere
Judges and reviewers that are themselves measured, verse-level audits of published translations, inherited versus introduced readings.
Starting a language from nothing
Survey what exists for a language before committing, and seed drafting from Open Bible Stories where no Scripture exists yet.
Built for translation organizations
Your team, your languages, your books — with the walls in the right places.
Roles that match the field
Reviewers see exactly their assigned language and review in-app; project managers add the file shuttle — offline review packets that work without internet — and USFM export.
Offline-first review
One HTML file carries the whole chapter to a reviewer with no connection; their decisions import back with conflict protection and a full re-check of every correction.
API included
A public /v1 API with per-organization keys — drafting and checking from your own pipeline, with the same per-verse evidence attached.