How this analysis was made

17,153 analyses across 1,189 chapters and 15 dimensions — themes, theological insights, cross-references, historical context, key events and figures, geography, cultural impact, practical application, linguistic insight, symbolism, prophecy, liturgical use and study questions — holding about 49,074 individual items between them. Also 10,093 topic pages, and every hashtag that links a chapter to one.

Not the verse text. That is the World English Bible, a published translation, reproduced as it stands.

17 prompts — one per dimension, plus two for topics. Each one asks for a Christian biblical scholar’s reading of a single chapter, from an evangelical perspective, drawing only on the biblical text: no commentaries, no apocryphal sources, no material from outside Scripture. Each prompt specifies the exact shape of its answer, and the answer is checked against that shape before it is stored.

The models were OpenAI GPT-4.1 and GPT-5 models. The prompts are versioned in this project’s source repository, and every analysis written since records which model wrote it and which version of which prompt it was working from.

Nothing is generated while you read. The whole corpus was produced offline, ahead of time; opening a chapter reads a database and calls no model at all.

The analysis was written against the New Living Translation (NLT). The text rendered beside it on every chapter page is the World English Bible — a different, public-domain translation, used because it is free to publish and we do not hold a licence for a modern one.

Two consequences, both worth knowing:

• Where the analysis quotes a phrase, the wording will often not match the verse printed above it. That is the translation difference, not an error in the verse text.
• In about 304 places the analysis quotes Scripture, and the model sometimes quoted from memory rather than from the text in front of it — so the wording may not match the NLT either. Treat any quoted phrase inside the analysis as paraphrase unless it carries a reference. Where it carries a reference — “Genesis 5:24” — that reference is a link to the actual verse, and the verse is what to read.

• Not a substitute for reading Scripture. The chapter is on the page above the analysis for a reason. If the two ever disagree, the chapter is right.
• Not pastoral counsel. It does not know you, your church or your situation, and it should not be the last word on a decision that matters.
• Not reviewed line by line. Every analysis is checked to be well-formed. No human has read all 49,074 items, and a language model writing confidently about the Hebrew of a verse can be confidently wrong.
• Not a substitute for commentary. The prompts deliberately excluded commentaries and the scholarly tradition. That keeps the analysis inside Scripture; it also means it carries none of two thousand years of careful reading.

Some chapters have no entry for some dimensions. Where that happens the section is simply absent — no placeholder, no “no data available”. Liturgical use and prophetic fulfilment are the two with real gaps: a genealogy in Chronicles has little liturgical history, and the honest answer was to say nothing rather than to invent something.

World English Bible, 2020 stable text edition. Public domain; “World English Bible” is a trademark of eBible.org.

Source: ebible.org/web/. The same attribution appears beside the text on every chapter page.