j.jevmanual.
OfficialChecked 2026-09-21·jev-1.13.0

Jev Noul Explained: Yes/No Probability from 0 to 1

Interpret Noul values, define yes/no criteria, and choose thresholds using the costs of mistakes.

On this pageWhat is Noul?Request and criteriaWhy no confidence field?Choosing a thresholdNoul vs yes/no ChoiceRefund and PII examplesWhen not to use Noul

What is Noul?

Noul asks one yes/no question and returns the probability of yes. Near zero means no is much more likely; near one means yes is much more likely. Around 0.5 means the model assigns roughly equal probability to yes and no.

0.5 does not mean “medium.” It represents uncertainty between yes and no for the proposition you asked.

Request and criteria

{
  "type": "noul",
  "instructions": "Does the customer explicitly request a refund?",
  "criteria": {
    "true": "The message asks for money to be returned",
    "false": "No refund request is present"
  }
}

The optional criteria clarify the meaning of yes and no. An angry customer is not necessarily asking for a refund. A quoted statement about someone else’s refund may not describe this ticket’s request. Include those edge cases in evaluation.

Why no confidence field?

The returned noul is already a yes/no probability. TypeSafe does not include a separate confidence property for Noul. Do not try to read one, substitute zero, and accidentally route every answer to an error branch.

Choosing a threshold

Select a threshold using labeled examples and the cost of false positives and false negatives. A “possible sensitive data” check may favor catching more cases for review, while automated tagging might tolerate different errors. Thresholds do not override access control or policy.

An educational workflow might route intermediate probabilities to review. Keep the original probability in the review record; converting everything to a boolean hides ambiguity and prevents later threshold analysis.

Noul vs yes/no Choice

Both can express a binary decision, but their response shapes and interpretation differ. Use Noul when the probability of one statement is the useful output. Use Choice when you need an explicit category taxonomy such as supports, contradicts, and insufficient.

Refund and PII examples

For refund detection, identify an explicit request and attach a workflow flag. Do not issue money. For PII, a Noul judgment can trigger additional checks or review, but deterministic detectors should handle recognizable emails, account numbers, or other exact patterns.

When not to use Noul

Do not ask “how urgent?” or “which department?” as one Noul. Do not ask an unbounded extraction task such as “what is the email?” Use a parser to collect candidates and Choice to select among them, then validate and normalize in code.

Search the manual