Glossary: Threshold
Threshold is any numeric cut your service applies: Choice/Score confidence, Noul noul (or |noul − 0.5|), or a Score expectation (“past level 1”). TypeSafe publishes example floors; they are not defaults. A threshold is not a request field in the public evaluation body we read. It lives in the decision contract next to instructions and criteria.
This glossary entry is for builders who already heard of Jev / System One and need a citable definition before they reuse one τ everywhere. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Confirm confidence examples on docs.typesafe.ai. We do not sell or issue API keys.
How it shows up
Store thresholds next to the question id and the pinned model. Changing jev-latest under a fitted cut is a silent policy change — pin jev-1.13.0 (vendor snapshot: that is what the alias currently resolves to). Do not reuse a Noul τ on a Choice. Do not ask the model to “apply the 0.85 rule” — apply it in code.
Confidence is derived from the probabilities map on Choice/Score. Calibration is group-level under the RLCD vendor claim, not a per-row guarantee. Noul has no confidence. Score score may be fractional; TypeSafe’s jaggedness note says jev-1.13 is weak at treating that expectation as a physical quantity — threshold “past level 1,” do not reconstruct dollars.
The cut produces a verdict: act, review, or abstain. Schema-safe ≠ factually correct. A confidence threshold is the special case that talks only about Choice/Score peakedness.
Contrast
Threshold vs confidence threshold. Confidence threshold is one species. Threshold also covers Noul τ and Score-position cuts.
Threshold vs calibration. Fitting τ is operations. Calibration is whether the numbers match frequencies.
Threshold vs Score score. The score is position on a rubric. Confidence is peakedness. Gate both when stakes are high.
Threshold vs guardrail. A guardrail uses thresholds; it is not itself a number.
Common mistakes
- One global τ for every question and every side effect.
- Copying Noul τ onto Choice confidence.
- Asking Jev to apply the rule in
instructions. - Leaving τ next to
jev-latestafter you fitted it. - Treating TypeSafe’s 0.5 / 0.9 examples as shipped defaults.
See also
FAQ
Is there a threshold field on the request?
Not in the public evaluation body we read. Thresholds live in your service.
Score vs confidence threshold?
Score score is position on a rubric. Confidence is peakedness. Gate both when stakes are high.
Can I store τ in TypeSafe?
Not as a documented evaluation field. Store it in your config next to the question id and model pin.
Do illustrations of 0.85 bind me?
No. They are examples. Tune on labeled traffic.
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.