Jev Paper Judge

Whether a paper is easy to follow is a snap judgment, not a chain of reasoning. A reader in the field knows within a paragraph whether the terms are explained, whether the sentences land, and whether the point comes through. Jev is a System One model built for exactly that kind of fast, calibrated judgment: it answers a small, literal question with a probability instead of prose. That is why a whole paper is judged in seconds, and why the questions below are small.

1. Extract. pdf.js reads the PDF in your browser. Headings, the abstract, and the references are located, and the body is cut into passages of about a thousand words.
2. Ask small questions. The document structure goes to this site's server in one request. The server sends every passage to Jev in parallel, each with the same fixed set of questions: are sentences direct, are terms and symbols defined, does the passage signpost its purpose, does it explain its method, setup, or results. The title, abstract, introduction, and conclusion get their own set. Each answer is a calibrated probability, returned in well under a second.
3. Combine in code. The server turns the answers into weighted averages you can inspect. Anything countable, such as undefined acronyms or figures that are never referenced, is checked by plain code. Answers where the model was unsure are marked with a ?.

The result measures presentation, not correctness. A wrong paper can still be clear and complete.