Writing and improving Jev programs

Jev is a judgment model. It reads one state, answers every question in the request independently and in parallel, and returns a probability distribution over the answers you defined. It does not reason in steps and it does not generate text. Code owns the control flow, the arithmetic, and the policy. Jev owns the snap judgments.

A good Jev question is one a knowledgeable person answers in a second, given the right context. "Does this message convey urgency?" fits. "Analyze this message and decide what to do" does not fit. Break that task into small questions and compose the answers in code.

This guidance applies to jev-1.13. The docs mark several limits as likely to improve in later versions, so recheck the jaggedness page when the model changes.

Workflow

1. List the decisions your code must make. Write each as a branch, a threshold, or a ranking.
2. Write one question per judgment. Split any question that weighs two properties.
3. Pick the primitive whose answer your code acts on directly.
4. Build the smallest state that answers every question. Compute in code whatever code can compute.
5. Put every question that shares the state into one request, including questions that matter for only some inputs.
6. Combine the answers in code: branches, weights, and confidence gates.
7. Test against labeled examples. Read probabilities on the misses, then revise one or two questions at a time.

Choose the primitive