Skip to content

Glossary

Noul question

A noul question is a yes/no check: Laya returns the probability that a statement about your text is true, such as 'the customer asks for a refund'.

Swiss-hosted inference. Nothing you send is ever stored. Swiss data residency

Definition

What Noul question means

A noul question is Laya's binary primitive, inherited from the Jev protocol: ask whether a statement holds for the state and get back a probability. "Noul" is Jev's name for this type; think of it as a calibrated boolean.

Shape

criteria is optional. If present it is a dict with true and/or false descriptions:

Show technical details· json sample
json
"is_phishing": {
  "type": "noul",
  "instructions": "Is this email a phishing or scam attempt?",
  "criteria": {"true": "phishing, scam, or fraud", "false": "a legitimate email"}
}

Internally it is always rendered as two option markers, false: … then true: …, with defaults "no, the statement does not hold" and "yes, the statement holds".

The answer is noul (the probability of true, 0 to 1), confidence = max(p, 1 − p) (never below 0.5), and an action block. Threshold on noul directly: flag at 0.5 for balanced costs, lower for recall-critical checks such as threats, higher when false positives are expensive.

Known issue: label following

The model card documents that noul can follow its option labels rather than the state, most strongly on the English checkpoint: the "false:"/"true:" pair can dominate and return a confident "no" for clearly positive input (issue #156). Check noul answers on your own data. If they look stuck, ask the same thing as a two-option choice question with neutral keys:

Show technical details· json sample
json
{"type": "choice", "instructions": "Is this review positive?",
 "criteria": {"A": "yes, the review is positive", "B": "no, the review is negative"}}

On the fine-tuned typed-decisions benchmark, noul is the strongest primitive (0.857). See choice, score and noul.

Get started

Try it on your own text

Send any text with your own typed questions and get a calibrated probability for every answer. 6k free tokens, no card needed; then $0.0357 per 1M input tokens — 15% below Jev.

Swiss-hosted inference. Nothing you send is ever stored.