Primitive guide
Choice, Score, and Noul
Jev is TypeSafe's System One model. It is not a chat model. You send a state plus typed questions, and the reply is structured data your code can branch on. It does not write a paragraph. This page is JEV Hub's restatement for builders. It is not TypeSafe documentation. The official introduction remains the source: docs.typesafe.ai/introduction.
Same state, separate questions
One request can mix Choice, Score, and Noul. They are judged against the same state, separately, in parallel. Adding questions does not make them share context.
Choice picks from a list and returns a probability per option, plus confidence. Score rates the state on a rubric and returns a score, probabilities, and confidence. Noul answers whether a statement is true, as a probability from 0 to 1. Noul does not add a separate confidence field.
You can run a request on this site: the Jev Playground sends it to https://api.typesafe.ai/v1/systemone with your own TypeSafe key and shows the live answer.
One narrow question
Each question should ask one narrow thing. If a judgment has several factors, ask each factor separately and combine the results in your own code.
A support ticket shows the split. Do not ask whether the ticket was handled well. Ask whether it names a product, whether it requests a refund, and whether the wording is urgent. Your code decides how those answers add up. When the weighting changes, change your code, not the question.
State vs questions
Official guidance puts material to judge in state and the judgments in questions. Question ids are keys in the response; write the full question in instructions.
Ask for a snap judgment a knowledgeable person could make with the right context. “Does this message convey urgency?” is in scope. “Analyze this and decide the best course of action” is a System 2 prompt; split it. Source: TypeSafe primitives.
The hosted request this hub validates is POST https://api.typesafe.ai/v1/systemone with Bearer auth and JSON fields model, state, and questions. Community snippets that add a mode field are not the official shape.
Choice
Pick one named option from a closed set. Official docs: add an other option when the list may not cover every input. Returns choice, probabilities, confidence.
Use Choice when options have names but no order: billing vs technical vs other. Official docs recommend an other option when the list may be incomplete. choice is the selected option; probabilities covers every option; confidence summarizes how peaked that distribution is.
Source: Choice primitive.
Score
Place the state on ordered levels you define. The returned score can land between two levels. Returns score, legend, probabilities, confidence.
Levels are ordered and numbered from 0. Three descriptions means scores from 0 to 2. The returned score can sit between levels. Threshold in application code, not in the prompt. legend repeats the levels by number.
Source: Score primitive.
Noul
Ask a yes/no question. The probability is the signal; there is no separate confidence field. Returns noul (probability 0 to 1).
Near 1 is a strong yes, near 0 a strong no, near 0.5 uncertain. Optional true/false criteria can spell out what counts. There is no separate confidence field on Noul answers.
Source: Noul primitive.
What this hub will not claim
JEV Hub does not invent probabilities or treat community playground fields as official. Answers in the playground come back from TypeSafe unchanged. Model ids it accepts are the documented ones: jev-latest, jev-preview, and versioned ids such as jev-1.13.0. Pin a versioned id if you tune thresholds.
Continue in the Jev Playground, the Jev MCP server, or the workflow guides.