Essay
Jev and Claude as judges: 1,459 labelled cases
Cheap judgments, agent traces, and when sending uncertain cases to a stronger model pays off.
Writing
Guides from the work.
Guides explain what my tools do, how they work, and what to expect when you try them. Essays are working notes on tools and ideas I’m evaluating.
Start reading
Essay
We checked Jev’s calibration on 3,742 labelled decisions
A probability threshold is a claim about how often you will be right. On two public datasets, the claim was wrong in both directions.
Essay
What to use Jev for
TypeSafe’s System One model answers typed questions in a fraction of a second. Six places that changes what you can build, and what to keep away from it.
Project guide
Set up your local evaluation stack
Try a guided example, build your own check, or continue an existing workflow with your coding agent.
Project guide
Keep a lesson from your agent work
How Lessons learned captures corrections, combines repeated evidence, and proposes lasting instructions.
Project guide
Check the evaluator, too
Start with one quality question and distinguish development feedback from independent human validation.
Project guide
Inspect a skill before you use it
Read the findings from a static scan and record a reviewed version so later changes are visible.