- Ask your PDF a question—let Jev rank the source text that answers it. - Powered by jev-latest, the demo evaluates extracted text chunks through TypeSafe’s system_one API. - Every answer pairs the original text with a probability meter and visible evidence on the page. - Evaluate selected pages, compare alternative matches, and inspect the red highlights. - This is source-based answer selection: model probabilities show ranking confidence, not benchmark accuracy.