Before you press play
What you’ll find in the video
- Use Jev for a check with known outcomes. In this Rhesis integration, it evaluates categorical metrics; it does not generate tests or handle numeric metrics.
- Turn the evaluation into a Choice question. The input and reply become state, the prompt and steps become instructions, and the metric's categories become answer options.
- Decide whether you need a written explanation. The airline example distinguishes approved airlines, no airline, and another airline; Jev returns the selected category and probability without generated reasoning.
Worth knowing
Product demonstration, not an independent accuracy benchmark. The presenter connects Jev through OpenRouter and reports roughly twenty tests, with two failures for each of the two compared metrics. Those counts do not establish a general error rate or a complete run audit. Jev's categorical-only role here is a Rhesis integration limit, not a claim that Jev has no Score question type. Speed and cost comparisons are creator claims; no benchmark or application was run for these notes.