system-one-bench

What it is

A public catalog of experiments measuring what Jev can help an agent do, for individual agent users and builders. It records each experiment's measurement, baseline, and conclusion limits.

How it uses Jev

Jev makes decisions like choosing a palette, selecting items, ranking context, checking claims, picking tools, and triaging logs. The state is the task context; Jev returns calibrated probabilities used to select or rank options in code.

Primitives:choicescorenoul

Technique worth stealing

Define typed questions over shared state to get calibrated probabilities for agent decisions.

Try it

See the practical guides and evidence records for reproducible decision benchmarks.

View on GitHub