Optional ReadonlyBaselineOptional ReadonlyConfidenceConfidence floor threshold in [0, 1] (defaults to 0.60).
Optional ReadonlyEvaluationOptional ReadonlyMaxMaximum acceptable fraction of predictions falling below ConfidenceFloor (defaults to 0.20).
Optional ReadonlyMaxMaximum acceptable mean absolute score delta between champion and challenger (defaults to 0.15).
Optional ReadonlyMaxMaximum acceptable drop in mean confidence from baseline to evaluation (defaults to 0.10).
Optional ReadonlyMaxMaximum acceptable p95 challenger-to-champion latency ratio (defaults to 3.0).
Optional ReadonlyMinMinimum acceptable label/band agreement rate between champion and challenger (defaults to 0.85).
Optional ReadonlyMinMinimum valid confidence observations required before flagging confidence degradation (defaults to 100).
Optional ReadonlyPairsOptional ReadonlyProblemProblem type for numeric score agreement ('classification' uses probability cut unless tolerance given; 'regression' uses tolerance band).
Optional ReadonlyScoreMaximum absolute difference between champion and challenger numeric scores counted as agreement in regression/tolerance mode.
Input to ComputeConfidenceAndChallengerDivergence.