jev-harness

How we reviewed this: we read the repository at the pinned commit and installed it in Docker. It ships examples and an offline eval CLI rather than a test suite, so there is no pass/fail tally to report.

What it does

jev-harness, by Antonio Coppe, is the layer between a probability and an action. Jev gives you a choice, a score or a yes/no with a confidence; this adds the parts every team writes afterwards:

• a policy that maps answers to actions;
• a confidence gate, so low-confidence answers escalate rather than act;
• shadow mode, to run a policy alongside production without doing anything;
• recipes for the shapes that recur;
• an offline eval CLI to run a policy against recorded cases.

That list is the right list. Shadow mode in particular is the feature most teams skip and then wish they had.

What to know

It’s early — three stars, eleven commits — and the repository has examples rather than tests, so treat the API as unsettled and read the source before depending on it.

And it doesn’t change what you send: that’s still your application’s decision.

Verdict

A good inventory of the missing pieces, and worth reading even if you write your own. If you’re about to build a confidence gate by hand, read this first.

For the same layer inside a specific tool, see the policy gate in Jev-cu and the bands in Abide.

See how it compares with other tools in Best Jev tools, tested hands-on.

Review updated Sep 20, 2026. Numbers quoted from the project are its author's own; we don't publish our own measurements of Jev.