vigiles API reference
    Preparing search index...

    Interface EvalMeasurements

    The five measurements an eval file may declare — the runners that already exist, keyed by their own names, so migrating is renaming a call to a key. EXACTLY ONE must be present: zero is a file that declares nothing, and two is a file whose author expected both to run.

    interface EvalMeasurements {
        measure?: MeasureSpec;
        measureArms?: ArmsMeasureSpec;
        measureSelectionMatrix?: SelectionMatrixSpec;
        measureTriggerRate?: TriggerRateSpec;
        runEval?: EvalSpec<Metrics>;
    }
    Index

    Properties

    measure? measureArms? measureSelectionMatrix? measureTriggerRate? runEval?

    Properties

    measure?: MeasureSpec

    Checks scored over N trials of one task — paid_measure.

    measureArms?: ArmsMeasureSpec

    The same checks per arm — paid_measureArms.

    measureSelectionMatrix?: SelectionMatrixSpec

    The N×N skill-selection collision matrix — measureSelectionMatrix from vigiles/claude-code. Declared here because a description is a description whatever harness answers it; only eval-entry.ts resolves it to a runner.

    measureTriggerRate?: TriggerRateSpec

    Does a skill's description actually fire — paid_measureTriggerRate.

    runEval?: EvalSpec<Metrics>

    A/B across arms with derived metrics — paid_runEval.

    Settings

    Member Visibility

    On This Page

    Properties