Jev, in context.

What the results mean

These are recorded API responses from Jev, evaluated against the exact case text available under each experiment. Switching a scenario replays its result. It does not make a fresh API call.

The examples draw on a sample of your Traces history, local preferences, scheduled-job metadata, and recent writing and teaching projects. They are not a comprehensive review of your history. Private source details have been generalized; comparison messages, draft passages, and listings are invented examples.

The session review uses an assistant-prepared evidence summary, including your feedback. That framing can affect the result. A model judgment is another signal to inspect, not an independent proof of success.

Choice probability compares the possible answers. Confidence describes how concentrated that distribution is. Neither substitutes for evidence or your authorization. The review threshold is an illustrative policy that has not been calibrated on your data.