Research mode

Normally SourceText answers only from the text it can retrieve. Research mode lets it also tell you what commentators, traditions and scholarship have said — clearly marked as that, and never mixed in with the text. You decide when to leave the text. It never decides for you.

What changes

In research mode the answer may report things it cannot look up: how a passage has been read, what a tradition holds, what the historical setting is generally taken to be, where scholars disagree. None of that comes from the corpus, so none of it can be cited from the corpus — and the answer says so in the sentence rather than leaving you to guess.

You will see wording like “Augustine is generally read as holding…” rather than “Augustine says…”. That is not vagueness. It is the honest form of a claim that comes from memory instead of from a page, and it is the difference between a report you can go and check and one you cannot.

What does not change

• Anything the answer says the text says is still retrieved and still cited. Research mode adds a second kind of statement; it does not loosen the first.
• Every claim is still marked for what it is — asserted by the text, inferred from it, or open.
• It still will not adjudicate. Where interpreters disagree, you get what each one holds, in its own terms. Deciding between them is your work, not the tool's.
• It will not quote from memory. A remembered quotation is the easiest thing in the world to get subtly wrong, so what you get is a paraphrase that says it is one.

Some commentary is held, and that changes the answer

SourceText holds the Jamieson-Fausset-Brown commentary (1871, public domain). Held means retrievable: ask what JFB says about a passage and you get the actual paragraph, cited, marked as commentary and ranked below the text — in the default mode as much as in research mode. Nothing about that is a guess, so nothing about it needs a hedge.

That is the difference research mode is really about. It covers the sources SourceText does not hold. Where a commentary is here, the answer goes and reads it rather than remembering it — and the instruction to do so is written into the discipline, not left to luck.

A commentary is never mixed into the text. Ask for a passage and you get translations; the commentary comes when you ask for it by name. Three chapters are missing from the copy we hold — Psalm 146, Mark 15 and Hebrews 13 — and the answer will tell you so rather than filling the gap.

How to read a research answer

A research answer is badged as one, and set apart from the answers around it, because a conversation can move between the two modes freely — the badge belongs to the message, not the session. When you look back at a transcript, or send someone a screenshot, the answers that left the text are the ones wearing the badge.

Anything unverified is somebody's report of a source, filtered through a model's memory of it. Treat it as a lead worth following, not as a finding. Where a claim matters to you, go to the source.

When to use which

Use the default when the question is what does this passage say. Use research mode when the question is how has this been read, or when you have met a claim somewhere else and want to know what the text actually does with it. Real study runs in both directions, which is why the switch sits next to the question and applies to that question alone.

Where it is available

Research mode is offered on the Bible corpus. It is not offered on the Danish or EU law corpora, deliberately: the sources a legal question would reach for — case law, forarbejder, commentary — are exactly the ones an AI is most likely to invent a convincing reference for, and a citation that cannot be looked up is worse than no citation. Those corpora answer from the statutes they hold, and say so when a question goes beyond them. That will change when SourceText holds legal secondary sources it can retrieve rather than recall.