Benchmarks & research

jev-agent-failure-benchmark

What it does

This benchmark compares Jev with a strong language model at attributing agent failures in the text subset of Who&When Pro.

How you can use it

Find out who caused a failed AI task, at which step, and what went wrong. Have a developer try this test's small sample and cost estimate before asking Jev, an AI service, to judge the failures. The mistakes were added deliberately, so this is practice for investigating failures, not proof it can explain real incidents.