How to read Jev probability distributions
Every Choice and Score answer includes probabilities. Choice keys are your option names; values are floats that sum to 1. Score keys are level indices as strings ("0", "1", …); score is the probability-weighted position across those levels. Noul has no probabilities map.
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.
The distribution is the raw object. Confidence is a collapse of its shape (peaked vs flat). If you only store the winner, you cannot audit abstain bands, dual-queue routing, or a later change to your own statistic.
How to use
choice: "technical" with { technical: 0.51, billing: 0.49 } is a different system event from { technical: 0.98, billing: 0.02 }. Both can produce the same winner. Confidence is meant to distinguish them. If your domain cares about the runner-up (for example dual-queue routing), read the map.
score is already the expectation over levels. You can recompute it from probabilities and legend for logging. Do not treat a fractional score as a physical measurement; see the numeric-calibration warning on jaggedness. Keep arithmetic in code. Do not treat score as dollars.
TypeSafe says the full map is always there on Choice and Score. You may compute another statistic than official confidence. You cannot replace their field in the request. Ranked routes that consume the ordered mass belong in your code — ranked route.
Store the versioned model, the full probabilities, and confidence. If an alias moves, you will want to know which ID produced the distribution you tuned against. Persist that JSON on the decision trace.
When not to use (or misread) the map
- Do not look for
probabilitieson a Noul. Noul is P(yes) only. A value of 0.5 means uncertain, not “medium,” and there is no confidence field. - Do not assume two Nouls will form a complementary pair that sums with this map. Different questions, different answers.
- Do not reconstruct a dollar or a minute from Score mass. Threshold the expectation; do not interpolate a quantity.
- Do not drop the map in logs “to save space” if you later need to explain an abstain. You cannot audit bands without it.
- Do not invent search-volume stories about how often 0.51 vs 0.98 appears. Measure on your traffic.
If TypeSafe documents that Choice probabilities always sum to 1 and you see otherwise, log the raw body and ask them. This site does not invent a second evaluation route; first-party evaluation is POST /v1/systemone.
Production notes
Confidence-gated routing still starts from this object — confidence-gated routing. A signal in our glossary is what you chose to persist, not a hidden TypeSafe channel. Label vendor latency or price figures if you print them next to a distribution.
Hub: Primitives. Official confidence page: docs.typesafe.ai/confidence.
See also
FAQ
Do Choice probabilities always sum to 1?
That is what TypeSafe documents. If you see otherwise, log the raw body and ask them.
Can I use another statistic than official confidence?
Yes. They say the full map is always there.
Ranked route?
Does Noul include this map?
No. Read noul only.
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.