Member Junction
    Preparing search index...

    What InferProcessor.UpsertFeatureValues writes for one record; the key parts (pin, input hash) are required.

    interface FeatureValueWriteParams {
        AIPromptRunID?: string;
        Confidence?: number;
        ConstraintHash?: string;
        Context: RecordProcessorContext;
        FeatureValueCacheID?: string;
        InputHash: string;
        OmitValues?: boolean;
        OutputConfidence?: Record<string, number>;
        Payload: unknown;
        Pin: FeaturePipelinePinData;
        PromptID?: string;
        Reasoning?: string;
        Record: RecordRef;
    }
    Index

    Properties

    AIPromptRunID?: string
    Confidence?: number
    ConstraintHash?: string
    FeatureValueCacheID?: string
    InputHash: string

    SHA-256 of the record's normalised inputs (its cache key hash), part of the Feature Value key.

    OmitValues?: boolean
    OutputConfidence?: Record<string, number>
    Payload: unknown

    The pipeline's payload the outputs are read from (by each output's Ref); unknown at that boundary.

    The pin that computed the values: its prompt version hash, primary model and parameters hash are part of the key.

    PromptID?: string
    Reasoning?: string
    Record: RecordRef