apiVersion: nines.dev/v1alpha1
kind: ServiceLevelObjective
metadata:
    name: agent-evaluation-pass
    project: production
spec:
    lifecycle: experimental
    measurement:
        type: availability
        metric: agent_evaluations_total
        labels:
            service: agent
        successLabels:
            verdict: pass
        countFrom: counter
        scrapeIntervalSeconds: 15
        observationPoint: All tasks selected for the declared evaluation cohort with terminal verdicts. Pin rubric, evaluator and sampling policy using bounded labels. Failed and unevaluable verdicts stay in the denominator. Track evaluation completion coverage separately.
    source: primary
    displayName: Agent outcome evaluation
    service: agent
    target: 0.99
    windowSeconds: 2419200
    stepSeconds: 60