JEV / DECISION ENGINEERING
What is Jev? Typed decisions explained with a support ticket
Jev reads state and predefined questions to return typed answers. Learn the suitable tasks, limits, and what code must still do.
A concrete example
A customer writes: “My package says delivered, but I have not received it.” The application passes that message as state and asks which of shipping, billing, technical support, or other should handle it. Jev chooses among those options and returns probabilities. The application retrieves orders, routes the ticket, and records the action.
Three question types
Choice selects among fixed options. Score rates against a defined scale. Noul returns the probability that a statement is true. All three can be asked about the same state in one request.
The boundary
Jev does not write a customer reply, summarize free text, or invent tool arguments. It cannot know order facts you did not provide. A valid output can still be wrong, so include an other option, thresholds, and review.
When is Jev worth evaluating?
Ask three questions: Is the output naturally a finite option, rubric level, or statement probability? Can the required facts be obtained and verified? Does code have a safe path for uncertainty and failure? If you need a letter, summary, code, or unavailable facts, use an appropriate tool or text model instead.
Multiple answers do not make one business decision
A support ticket can ask for the first team, urgency, and tone in one call. Jev returns independent judgments. Your application defines how to compose them, which permissions apply, and how failures are handled. Each question can then be evaluated separately.
The shortest useful learning path
Work through the ticket below by hand. Try de-identified text in the official Playground. Use this site's lab to draft a request, then test on your labeled cases. Our lab is a local draft tool and does not call Jev.
Progress is stored only in this browser