Primitives
Choice, Score, Noul, confidence, and how questions compose in code.
Choice vs Score vs Noul
Three question types, one endpoint. Choice picks an option, Score places a state on a rubric, Noul returns P(yes).
Jev Choice questions
Choice selects one option from a map you define (up to 255) and returns the winner, a full distribution, and confidence.
Jev Score questions
Score rates a state on 2–10 ordered levels and returns a probability-weighted value that can land between levels.
Jev Noul questions
A Noul returns P(yes) for a yes/no question. There is no separate confidence field — the probability is the signal.
Jev confidence vs probability
Choice and Score include a confidence statistic derived from the distribution. Noul does not. Thresholds should scale with risk.
Writing Jev criteria that the model can use
Criteria are the answer space: a map for Choice, an ordered list for Score, optional true/false glosses for Noul.
How to read Jev probability distributions
Choice and Score return a full distribution. Use it when confidence is too coarse or you need expected values.
Structured instructions and field paths
Instructions can be strings, objects, or arrays. Point at state with backtick paths such as ticket.messages[0].text.