What is a System One decision model?

Updated 24 Sep 2026

A System One decision model maps an input into a defined decision structure. Instead of requesting an open-ended explanation, you specify the options or scale your application can use. The format is constrained; the judgment can still be wrong.

What do Choice, Score and Noul mean?

In Jev, Choice selects a label, Score evaluates an ordered scale, and Noul estimates a binary proposition. The API calls the input state. Separate questions are evaluated independently, so do not expect one answer to depend on another answer in the same request. See the official introduction.

Example: route a customer request

For “Can someone give me a price for repairing a fence?”, define quote, schedule, complaint and other as the available labels. A quote decision could open an estimating queue. This is an illustrative scenario, not a recorded model result.

Validate the returned label and confidence before acting. Send malformed responses, unfamiliar requests and uncertain decisions to review. Our fallback routing recipe demonstrates that policy.

Does confidence prove correctness?

No. Evaluate error rates on representative labeled examples. A confident wrong label is still wrong. Keep consequential actions behind checks appropriate to their impact.

When should I use a text model instead?

Use a generative model when the result needs prose, a detailed explanation or a draft. A typed decision can choose which workflow runs next. The term System One describes a product approach here; it is not evidence that a model thinks like a person.

Sources and test status

Reviewed 24 September 2026 against the Jev quickstart. This guide explains documented behavior; we have not independently measured model accuracy or latency. For deployment tradeoffs, read Jev or Laya: how to choose.