JevGPT

How it uses Jev

For each output word, Jev chooses among buckets of a 20,000-word vocabulary, then a runoff decision picks the next word; the loop stops on its completion judgment or a word limit.

Why it's interesting

It deliberately pushes Jev against its intended boundary: typed choices can be chained into text, but the bucket runoff and repetition controls expose why decision models are awkward language generators.

How often does Jev run?

The project's Jev call cadence has not been verified from public evidence.