A/B testing should reduce uncertainty rather than manufacture confidence. We define hypotheses, metrics, guardrails, readiness, instrumentation, and an appropriate analysis plan before interpreting a result. We do not guarantee uplift or declare a winner without adequate evidence.