Four numbers and labels often collapse into one vague “confidence” badge. Keeping them separate makes a typed decision inspectable.
Selected-option probability
For a choice, Jev can return a distribution across your named candidates. The probability attached to the selected candidate is one signal. For a boolean, the probability estimates the proposition being true.
Provider confidence
TypeSafe confidence is separate metadata for choice and score questions. It summarizes how concentrated the distribution is. Boolean answers do not have this same confidence field.
Rubric position
A score is a probability-weighted position on an ordered, zero-indexed rubric. A score of 2.6 is not 60% certainty and should never be rendered as generic confidence.
Application policy
Your code can require both selected probability and confidence to cross thresholds, or route all external side effects to review. Changing these thresholds should not call the model again. That is why Route Lab shows “model choice” and “application path” separately.