Decidedness and confidence

A choice reading has a tier and a decidedness.

• Tier: the verdict. It is the tier with the most probability. A tier’s mass is the sum of its options’ probabilities.
• Decidedness: whether the verdict is stable. It compares the top tier’s mass with the runner-up tier’s mass.

Undecided

A reading is undecided when the top tier leads the runner-up by less than margin. With two tiers, margin: 0.2 makes a top tier under 0.60 undecided.

Undecided is not a severity. It says the reading does not decide, not how bad the code is. The response is to recalibrate the rule, not to act on the code. See calibration.

Set margin per rule, from how far that rule’s readings move between reruns.

The provider’s confidence

A score or choice answer can carry confidence: the provider’s own number, 0 to 1. jint does not derive it from the probabilities, and no verdict reads it.

jint reports it in --json and on the run view’s answer card. It is absent when the provider returns none.

CI

An undecided reading must not fail a build, because its tier can change between runs. A tier change needs a near tie, and a near tie is an undecided reading. A build that fails only on decided readings does not flake.

Limits

• margin: 0.2 is an example, not a fitted value. Choose a margin per rule from its own reruns.
• The classifier has no default threshold. Set one per question.