Confidence Model

Governance source of truth for how revision confidence is calculated and displayed across recipes.

Model Summary

Confidence combines weighted revision votes with real cook feedback. As more cooks report outcomes, that feedback has more influence on the score.

Baseline prior confidence starts at 55% and is then adjusted by evidence.

Primary signal is weighted review outcomes.
Cook outcomes add evidence over time: 'no notes' raises confidence, while 'have notes' lowers it.
Stable, uncontested revisions can gain confidence over time.
Active pending contention reduces confidence until resolved.

Maximum stability bonus: +12 points when a revision remains uncontested with strong consensus.

Maximum contention penalty: -10 points while active pending revisions indicate unresolved disagreement.

Maximum cook-outcome contribution: 24% of the final score once enough cook evidence accumulates.

Confidence bands shown in UI: