Jev Noul questions
Noul is the yes/no primitive. The answer is a single float noul from 0 (no) to 1 (yes). Near 0.5 means the model is not committing. Official docs are consistent: a Noul answer has no confidence and no probabilities map. Reading answers.foo.confidence on a Noul is undefined.
This page is for builders who already heard of Jev / System One and want a production-minded mental model. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Confirm facts on docs.typesafe.ai. We do not sell keys.
Noul is P(yes). That is the whole signal. Do not translate 0.5 into “medium,” “amber,” or “somewhat skilled.” Those words belong on a Score whose levels you wrote. Confidence is peakedness of a Choice/Score distribution; Noul has none.
How to use
Write a statement you can threshold. “Does the resume mention distributed systems?” is a Noul. “Is this candidate strong in Python?” is usually a Score, unless you define “strong” as a crisp condition you would ask a colleague as yes or no.
Optional criteria align the labels with the instruction:
"criteria": {
"true": "Explicitly time-sensitive",
"false": "No urgency expressed"
}A Noul where true means no and false means yes is a documented failure mode. Keep the glosses a person could apply without flipping polarity.
Threshold in code: auto-act above a high bar, abstain near 0.5, auto-deny below a low bar — or the reverse, depending on fail-closed. Bands are yours. Do not invent a confidence field on the wire. Many Nouls in one POST /v1/systemone is the point of fan-out; compose atomic questions in code.
Use Noul for screens: refund requested, injection-like instruction present, PII mentioned, policy clause triggered. Then let Choice pick a queue if you still need an unordered set.
When not to use Noul
- You need ordered severity or skill. Use Score. A Noul of 0.5 is uncertain, not medium.
- You need one winner among many labels. Use Choice, and add
other/noneif the set is incomplete. - You need a peakedness statistic. Noul does not return confidence. Distance from 0.5 is the closest analogue, and you compute it.
- You want two Nouls to obey
P(A) + P(not A) = 1. The jaggedness page shows thatP(refund)and1 - P(not refund)need not sum to 1. Do not enforce arithmetic identities the model does not guarantee. - You are using Noul as a score out of 100, a dollar, or a star rating.
TypeSafe also publishes a self-consistency cookbook for nouls. Treat that as vendor guidance, not a warranty.
Production notes
Gating should use the noul, not a missing confidence field — see confidence-gated routing. Log the float, the pin, and your band. Store a decision trace so a reviewer can see why you auto-acted at 0.91 and held at 0.48.
Hub: Primitives. Glossary: Noul. Official: docs.typesafe.ai/primitives/noul.
See also
FAQ
Is 0.5 a medium Score?
No. Yes and no are equally likely.
Can I add confidence myself?
You can define bands. Do not invent a field on the wire.
Many Nouls in one call?
Yes — that is the point of fan-out.
Does Noul return a probabilities map?
No. The single float is the distribution.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.