Benchmarks & research

Jev calibration study

NXTG.AI compares Jev's probabilities with Qwen and Claude on contract questions and internal routing tasks, then tests adding retrieved context.

Bookmark Keep this in your collection.

Leave a noteWhat would you try with this?

Only you can see your notes.

Source screenshot of Jev calibration study — SOURCE SCREENSHOTFull screenshot ↗

What it does

Public contract records support part of the study; internal inputs and labels remain private. Some routing labels are recorded agent assignments, not human judgments.

How you can use it

Your developer can borrow this method to check an AI tool's confidence against known correct answers. They can ask questions with and without similar past records. You could also test a simple vote of past examples. Gains vary by task. The study found no clear improvement for some questions. Original internal records remain private.