What it does
Checks Jev questions against labelled data and reports whether each can gate decisions, rank examples, or carries no useful signal.
Benchmarks & research
jev-calibrate
Bookmark: jev-calibrate
Keep this in your collection.
Leave a noteWhat would you try with this?
: jev-calibrate
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Start with support messages where you know whether the writer is frustrated. A developer can use this tool to compare the AI's answers with your own. It shows when the AI is confidently wrong. The examples go to an outside AI service, connected through an access key supplied by that service, not your personal password.