monad.ai
    Preparing search index...

    Type Alias WeightReport

    A point-in-time snapshot of learned scorer weights with change context.

    Returned by getWeightReport and exposed via GET /.mesh/weights.

    type WeightReport = {
        current: Record<string, number>;
        defaults: Record<string, number>;
        delta: Record<string, number>;
        health: WeightHealth;
        lastUpdatedAt: number | null;
        namespace?: {
            blended: Record<string, number>;
            current: Record<string, number>;
            delta: Record<string, number>;
            maturity: number;
            namespace: string;
            sampleCount: number;
        };
        stable: boolean;
        updateCount: number;
    }
    Index

    Properties

    current: Record<string, number>

    Current learned weights (same keys as DEFAULT_WEIGHTS plus any custom scorers).

    defaults: Record<string, number>

    Baseline values the system started from (hard-coded defaults).

    delta: Record<string, number>

    current - defaults per scorer; positive = reinforced, negative = penalized.

    health: WeightHealth

    Diagnostic health signals for the learning loop. See WeightHealth.

    lastUpdatedAt: number | null

    Unix millisecond timestamp of the most recent weight update, or null if never updated.

    namespace?: {
        blended: Record<string, number>;
        current: Record<string, number>;
        delta: Record<string, number>;
        maturity: number;
        namespace: string;
        sampleCount: number;
    }

    Namespace-local report when getWeightReport(namespace) is requested.

    stable: boolean

    True when no delta exceeds 5% of its default weight.

    A stable system has not yet learned much, or has converged back to near-default weights after a period of learning. Not necessarily a problem — a homogeneous mesh naturally converges to defaults.

    updateCount: number

    Total number of gradient steps applied since the daemon started (or since last reset).