Evalkeeper
    Preparing search index...

    Type Alias RunRecord

    A stored test: one suite on one model, with what it ran with and its metrics.

    type RunRecord = {
        datasetVersion: string | null;
        error: string | null;
        estimatedCostUsd: number | null;
        finishedAt: Date | null;
        graderVersion: string;
        id: string;
        maxOutputTokens?: number | null;
        metrics: RunMetrics | null;
        model: string;
        promptVersion: string;
        reasoningEffort: string | null;
        repeats: number;
        set: EvalSet;
        startedAt: Date;
        startedBy: string | null;
        status: RunStatus;
        suite: string;
        temperature?: number | null;
    }
    Index
    datasetVersion: string | null

    Changes when the cases or their answers do, so tests compare.

    error: string | null
    estimatedCostUsd: number | null
    finishedAt: Date | null
    graderVersion: string

    The suite's grader version when last graded.

    id: string
    maxOutputTokens?: number | null

    The output-token cap tested; null or unset: the call's own default. Kept like temperature: a test at another cap is out of date.

    metrics: RunMetrics | null
    model: string
    promptVersion: string

    Hash of the instructions it ran with (see instructionsHash).

    reasoningEffort: string | null

    The reasoning setting tested; null: the model's own default.

    repeats: number
    set: EvalSet
    startedAt: Date
    startedBy: string | null

    Who started it; null for a scheduled check.

    status: RunStatus
    suite: string
    temperature?: number | null

    The temperature tested; null or unset: none sent. Stores keep it, so a test at another temperature than the live one shows as out of date.