Member Junction
    Preparing search index...

    Computed confidence-floor and champion-vs-challenger divergence metrics.

    interface ConfidenceAndDivergenceMetrics {
        AgreementRate: number;
        BaselineMeanConfidence: number;
        BelowFloorRate: number;
        ChallengerDiverged: boolean;
        ChallengerFailureCount: number;
        ChallengerTimeoutCount: number;
        ConfidenceDegraded: boolean;
        ConfidenceInsufficientData?: boolean;
        DivergenceInsufficientData?: boolean;
        EvaluationMeanConfidence: number;
        MeanAbsoluteScoreDelta: number;
        P95LatencyRatio: number;
        PairCount: number;
    }
    Index

    Properties

    AgreementRate: number
    BaselineMeanConfidence: number
    BelowFloorRate: number
    ChallengerDiverged: boolean
    ChallengerFailureCount: number
    ChallengerTimeoutCount: number
    ConfidenceDegraded: boolean
    ConfidenceInsufficientData?: boolean
    DivergenceInsufficientData?: boolean
    EvaluationMeanConfidence: number
    MeanAbsoluteScoreDelta: number
    P95LatencyRatio: number
    PairCount: number

    Pairs with a champion output. Only these count anywhere: toward the sample size, the agreement rate, the delta, latency and the timeout and failure counts. The challenger side is an output, a timeout, an error, or nothing (a failure too).