Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up
primitive

Noul

A statement about the state, answered with a probability that it holds. Named after the Jev primitive; the threshold for "yes" is 0.5 unless the caller sets one.

On Metatext decisions, Jev leads with 0.851 mean accuracy against a 0.701 majority baseline.

Metatext decisions

Every model is zero-shot on this suite.

#ModelMean accuracyBaselineBeats baseline
1JevTypeSafe0.8510.7012/3
2GLiClass base v3.0Knowledgator0.8080.7012/3
3GLiClass large v3.0Knowledgator0.8050.7012/3
4GLiClass modern large v3.0Knowledgator0.8050.7012/3
5GLiNER2 largeFastino0.8020.7012/3
6GLiClass multilang ultraKnowledgator0.7840.7011/3
7GLiNER2 baseFastino0.7670.7011/3
8GLiClass instruct base v1.0Knowledgator0.7250.7011/3
9GLiClass instruct large v1.0Knowledgator0.7170.7011/3
10GLiNER 2.5 baseFastino0.6980.7011/3
11GLiClass edge v3.0Knowledgator0.6270.7011/3
12GLiNER 2.5 smallFastino0.6210.7010/3
13GLiGuard 300MFastino0.5850.7011/3
14LAYAConvai Innovations0.5720.7011/3
15Vonwfzyx0.5580.7011/3
16GLiNER 2.5 multiFastino0.5520.7011/3
17LAYA multilingualConvai Innovations0.5500.7010/3
18LAYA typed-decisionsConvai Innovations0.5480.7011/3
19GLiNER multitask v1.0Knowledgator0.5160.7010/3
20GLiNER2 multiFastino0.4540.7010/3
21OpenJEVopen-jev0.3830.7010/3

typed-decisions

Every model is zero-shot except LAYA typed-decisions, which was fine-tuned on this suite.

#ModelMean accuracyBaselineBeats baseline
1LAYA typed-decisionsConvai Innovations0.8570.6556/6
2JevTypeSafe0.7930.6555/6
3GLiNER2 largeFastino0.5320.6553/6
4GLiNER 2.5 baseFastino0.5250.6550/6
5Vonwfzyx0.5080.6550/6
6LAYA multilingualConvai Innovations0.4970.6551/6
7LAYAConvai Innovations0.4870.6551/6
8GLiClass large v3.0Knowledgator0.4370.6551/6
9GLiClass multilang ultraKnowledgator0.4330.6550/6

The questions

  • agent_traces.needs_human · noul · our tasks
    This step should be escalated to a human operator
  • customer_service.refund_requested · noul · our tasks
    The customer explicitly requests a refund or money back
  • security_incidents.actively_exploited · noul · our tasks
    There is evidence this issue is being actively exploited right now
  • td_agent_trace_observability.needs_review · noul · LAYA's public suite
    This trace requires human review.
    false · true
  • td_customer_service.needs_human · noul · LAYA's public suite
    This conversation requires a human agent rather than automated handling.
    false · true
  • td_invoice_processing.duplicate · noul · LAYA's public suite
    This invoice appears to duplicate an invoice already submitted.
  • td_invoice_processing.matches_order · noul · LAYA's public suite
    The invoice reconciles with the purchase order and the recorded delivery.
    false · true
  • td_security_incidents.credential_compromise · noul · LAYA's public suite
    The evidence indicates a credential or account has been compromised.
  • td_security_incidents.true_positive · noul · LAYA's public suite
    This alert reflects genuinely malicious or unauthorised activity.
    false · true

Frequently asked questions

What is a Noul question?

A statement about the state, answered with a probability that it holds. Named after the Jev primitive; the threshold for "yes" is 0.5 unless the caller sets one.

Which model is best at noul?

Jev leads on Metatext decisions at 0.851 mean accuracy over 3 questions.

Related

Choice

A question with a fixed set of named options and a description for each; the model returns a probability per option and the answer is the argmax.

Score

An ordered scale (none / low / medium / high, 1–5, …). The model returns a distribution over levels and the answer is the expected level. The primitive relevance classifiers cannot do.

Extraction

Return the text spans that match each label (person, organization, date, money, …). Jev and the pure decision models never return spans; only the GLiNER family does.