Glossary: Confidence threshold
Confidence threshold is a number you choose. If Choice or Score confidence is below it, code abstains or reviews instead of taking the argmax. TypeSafe examples often use 0.5 as a “genuinely unsure” floor and 0.9 for high-stakes acts. Those are illustrations, not shipped defaults. Noul uses a separate τ on noul, not confidence.
This glossary entry is for builders who already heard of Jev / System One and need a citable definition before they copy one floor onto every question. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Confirm the confidence page on docs.typesafe.ai. We do not sell or issue API keys.
How it shows up
Do not copy a Noul τ onto a Choice. Scale floors with irreversibility: showing a balance is not approving a transfer. Tune on labeled traffic; pin the model id you fitted. jev-latest currently points at jev-1.13.0 on the vendor snapshot we read; aliases move.
Confidence is a 0–1 summary of how peaked probabilities are. TypeSafe’s three-option demo uses (3 × largest probability − 1) / 2. They also say you may compute another statistic from the full map. Calibration is group-level under the RLCD vendor claim — not a per-row guarantee. Store the threshold in your decision contract, not in an undocumented API field.
This page is the Choice/Score species of threshold. The outcome is a verdict. Schema-safe ≠ factually correct.
Contrast
Confidence threshold vs threshold. Parent term includes Noul τ and Score-position cuts. This term is only the peakedness floor.
Confidence threshold vs Noul τ. Different statistic. Different field. Do not swap them.
Confidence threshold vs calibration. A cut can be stable while bins are miscalibrated on your slice. Measure both.
Confidence threshold vs ranked route. Rank without a floor is just argmax. The floor is what makes rank 1 optional.
Common mistakes
- One threshold for every question.
- Applying a Noul 0.85 to Choice confidence.
- Treating 0.5 / 0.9 as TypeSafe defaults.
- Fitting on
jev-latestand never pinning. - Asking the model to “only answer if confident.” Apply the cut in code.
See also
FAQ
Where do I store the threshold?
In your decision contract (code + config), next to instructions and criteria — not in an undocumented API field.
Can one threshold serve every question?
Usually no. Per-question, per-action floors are the production habit.
Does this apply to Noul?
No. Threshold noul or its distance from 0.5.
Is 0.5 official?
It is a common illustration for “genuinely unsure.” Fit your own.
Limits
This is a jev.pro working definition unless TypeSafe uses the same word as an API field. We do not invent search volume. Schema-safe answers are not automatically factually correct.
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.