Jev

Jev is a model returning typed probabilistic decisions instead of generated strings for automation workloads. Claims RLCD training and parallel sampler provide calibrated confidence, 70–500ms responses, and substantially lower costs on predefined-output-space tasks.