Jev Sandbox

Jev inverts the usual chatbot interaction. Instead of asking a model to write, you hand it a piece of text and a set of questions you wrote yourself, and it returns a typed answer per question: a probability for yes/no, a full probability spread for pick-one, or a weighted score across rubric levels you wrote. It never writes a sentence and cannot be asked to, which makes its uncertainty visible — the thing students are usually asked to read off a paragraph of fluent prose.

What you can do with it

The sandbox includes five student coaches and two faculty tools, a graph builder for your own judgments, and lenses that fetch live public data (news headlines, occupations, federal comments) to chart. A worked example judges an AI-use disclosure statement against a rubric and yes/no checks, which is a direct teaching case for AI Use and Disclosure Statements and for reading model probabilities critically. In the classroom it suits fast second readings of drafts against a rubric, triage of a stack of short responses, and teaching probability literacy.

Limits the project states itself

A vague question gets a confident wrong answer, so rubric wording carries more weight than with a chatbot; the model sees only the text in front of it and cannot check a source or notice sarcasm; and the text you submit travels to a company server for the length of the request, so it should carry no names or grades. Numbers like 0.97 will be taken as truth unless you teach students not to.

Notes

The site names no individual author; the source is a public GitHub repository updated in September 2026, written mostly in HTML with JavaScript and Python.

Connected Concepts

Formative Assessment, Feedback, Assessment Validity, Explainable AI, Trust Calibration, AI Use and Disclosure Statements