Every answer cites its source and labels its confidence — or admits it doesn't know.

Divine Data Answers responds to plain-language questions only from our corpus: the full 39-book Old Testament studied verse-by-verse, plus commentary, the modernized public-domain scripture text, and the catalog. Every claim carries the reference it came from and a tier label saying what kind of statement it is. Nothing is generated from a model's memory and dressed up as our work.

See how grounding works Where things stand

39Old Testament books studied verse-by-verse

23,145study verses

1,327commentary & analysis documents

111catalogued artifacts

Book and verse counts re-derived from the study files on 2026-08-03. The document and artifact counts are the last recorded database values and were not re-verified that day; the artifact count in particular has not moved since the catalog was first built, because catalog.json has not been regenerated, so it undercounts what the corpus now holds.

The grounded-answer contract

Five rules, enforced in code — not a promise in a marketing page

A generic chatbot will happily answer a question about Isaiah or the demiurge from its training data: fluent, unsourced, untraceable, and not our work. Divine Data is built the opposite way.

Built from the corpus only

Answers are assembled from data returned by the read-only corpus backend. An answer that leans on the model's own training to state a "fact" about the text is a violation, full stop.

Cited

Every claim carries the reference it came from — a verse ref, a document path, an artifact id — so you can check it against the source yourself.

Tier-labeled

Every statement is one of three kinds: plain-sense, anchored-with-citation, or a flagged interpretive extension explicitly labeled as ours. Never a fourth thing.

Honest about variants and gaps

Where sources differ, it shows the variants instead of silently picking one. Where the corpus doesn't cover something, it says so rather than filling the gap from memory.

Correctable

When we find we were wrong, we correct the record visibly. Our own eval write-up records a caught defect in the open rather than quietly editing it away.

—

Enforced, not aspirational

The contract is checked mechanically in the answer pipeline, gated by an internal truth-veto review, and regression-tested by a golden eval before anything ships.

What an answer looks like

The citation and the tier label are the interface

They aren't footnotes. They sit on every claim, styled so nothing reads as more authoritative than its tier actually is.

Your question

What does the corpus say about the suffering servant in Isaiah 53?

Plain-sense

What the source text says on its face, quoted from the modernized public-domain scripture text carried on the study verse.

Isaiah 53:5

Anchored with citation

A claim tied to a specific reference in the corpus — the verse-by-verse commentary at that location, retrieved this session, not recalled from training.

isaiah_study/ — verse commentary

Flagged extension — our lens

This project's Gnostic/Pistis Sophia reading, marked explicitly as our interpretation. It is never laundered into "fact" or into scholarly consensus.

labeled as ours

Honest limit

"The corpus does not address this." When a question falls outside the 39 loaded books and their commentary, the answer says so instead of inventing one.

Illustrative layout of the three permitted statement tiers. Claim text above is a description of each tier, not a generated answer.

The corpus

A specific, growing body of work — and we name its edges

All 39 Old Testament books, studied verse-by-verse through this project's Gnostic/Pistis Sophia lens, are loaded and queryable today. The study track continues, and the loader is book-agnostic — the corpus grows as books ship, including Phase B's wider shelf of Second Temple literature (1 Enoch and more), written but not yet loaded into the query engine.

Genesis
Exodus
Leviticus
Numbers
Deuteronomy
Joshua
Judges
Ruth
1 Samuel
2 Samuel
1 Kings
2 Kings
1 Chronicles
2 Chronicles
Ezra
Nehemiah
Esther
Job
Psalms
Proverbs
Ecclesiastes
Song of Solomon
Isaiah
Jeremiah
Lamentations
Ezekiel
Daniel
Hosea
Joel
Amos
Obadiah
Jonah
Micah
Nahum
Habakkuk
Zephaniah
Haggai
Zechariah
Malachi

Alongside the studies: commentary and analysis documents, the modernized public-domain scripture text carried on each study verse, the Pistis Sophia translation work where present in the corpus, and a catalog of project artifacts.

What Divine Data Answers is not

Not a generic AI oracle. It answers from this corpus. Ask it about the weather and it declines or redirects.
Not a model talking from memory. Parametric output dressed as our content is a violation, not a bug we tolerate.
Not religious authority, prophecy, or personal guidance. It reports what the texts and studies say, tier-labeled. It does not tell you what to believe.
Not a claim of scholarly consensus. Our interpretive lens is labeled as ours.
Not exhaustive yet. The queryable corpus today is the 39-book Old Testament plus the Pistis Sophia materials — not "the Bible," and not "all of Gnosticism." There is no New Testament study here yet, and Phase B's wider shelf isn't loaded into the query engine yet either.

Proof, not adjectives

What's actually built and measured today

Phase 1 is a working bring-your-own-key Claude query engine over the read-only corpus, with the grounded-answer contract enforced in code. Here is the recorded state of it.

Phase 1 golden eval — first fully-green live run, recorded 2026-08-01. Contract and refusal are hard gates.
Criterion	Result	Target
Grounding, citation coverage, tier labeling	19 / 19	100% — hard gate, met
Honest refusal (out-of-corpus questions)	7 / 7	100% — hard gate, met
Answer correctness (manual spot-check)	18 / 19 clean (~95%)	≥ 95% — met on that run
Offline test suite (no LLM calls, no database)	129 tests passing	Runs on every change

The correctness spot-check is not a gate you pass once: model output varies run to run, so every future live eval run gets its own fresh human review. On the recorded run it caught a real claim that overreached its citation — that finding is written into our eval record rather than quietly fixed.

Divine Data sells trustworthy access to a unique body of knowledge, not tokens.
— Divine Data strategy, the thesis in one line

Pricing

Included with the Library, or $3 a month on its own

Answers is a bring-your-own-key product: you bring your own provider key and pay that provider directly for inference, so our cost per answer is zero and there is no metering from us. The figures below are the current pricing we are running with; our CEO's sign-off is still required before anyone is charged. Full detail, including the free path and the accessible-pricing options, is on the pricing page.

Included

Answers with the Library

Already in every Library plan

$0 extraLibrary is $9/month, $90/year, or $180 once — Answers is included in all three, and in the $2 pay-what-you-can floor and free-on-request access.

Unlimited synthesized answers on your own key
Charging a second subscription for something that costs us nothing extra would be exactly the nickel-and-diming we bar

Current

Answers only

The query tool over the free corpus, without the gated study chapters

$3 / monthFor people who want the grounded query tool, not the study Library. Cancel in one click, effective at period end.

The grounded corpus gateway and multi-provider parity
Retrieval, citation, and tier-labeling on every answer
Saved threads and history
Not tokens — you pay your own provider for those

Free forever Corpus search and citation lookup stay free across the entire corpus, gated chapters included — you can always find a passage and cite where it is, without an account.

Not offered A managed tier where we supply the key is designed on paper but is not offered and carries no price, and we are not attaching a date to it. Until commercial terms are verified, caps are demonstrated to fail closed, and spend is approved, Divine Data is bring-your-own-key only.

Things we will not do

No fake "free trial" that silently converts to paid, no pre-checked upsells, no hidden auto-renew traps, no "call us to cancel."
No manufactured urgency, no bait pricing, no fake "free" tier.
No overage billing on any metered path — it fails closed, it does not bill past the cap.
No crippling the free path to force upgrades, and no pricing change that weakens the accessible path to grow a number.

Every figure above is quoted from the pricing page, which is the source of truth for what Divine Data costs.

Where things actually stand

The engine is built and tested. The web app is next.

There is no signup button here, because there is nothing yet to sign up to — and putting one here would be exactly the kind of thing we've promised not to do. What exists today is a working local bring-your-own-Claude query tool over the read-only corpus, holding the grounded-answer contract, with the eval results above on the record.

In the meantime, the written corpus is complete and readable now — no account, no card, no countdown, and the ebook has real checkout.

Read the corpus Read the eval results