These verification details describe how well a claim or result held up under review.

• Supported means the key points are backed by strong evidence or consistent agreement across the models.

• Mixed means some parts are supported, but other parts are uncertain, missing context, or depend on assumptions.

• Disputed means the models or sources conflict, or the claim could not be reliably supported, so you should not treat it as settled.

You can use these labels to quickly see what is solid and what needs more verification.

A claim is a single statement inside an answer that can be checked for truth or support. For example, Australia s capital is Canberra is a claim. SIBYL breaks answers into claims so it can verify important statements instead of trusting a long paragraph.

They are preset modes that trade speed for scrutiny.

• Light is quickest

• Fast is balanced for most use

• Deep is the most thorough and is best for high-stakes questions

In Fast and Deep, yes. Lite uses a smaller lineup for speed, but it still includes an independent

review step.

Plus includes a 2-week free trial, so you can try SIBYL before you pay.

It detects the type of problem you are solving and routes the workflow to models that tend to

perform best for that category.

A normal chatbot gives one answer. SIBYL generates multiple drafts, critiques them, verifies

key claims, and then produces one structured result that shows uncertainty and next steps.

The Trust Score is a 1 to 100 signal that gives you a quick summary of how reliable SIBYL

thinks the answer is. It is not a probability and it is not a confidence interval. It is a practical

quality score based on things like how well the models agreed, how many key claims held up

under review, and how strong the supporting evidence was.

A Veripoint is a single check inside of an answer. It captures one specific claim, what it was

checked against, and the outcome, such as supported, mixed, or not confident.

Yes. In Custom mode you can pick which models generate drafts and which ones review them.

SIBYL then combines the work into one final output.