From design to deployment, ISO 8800 defines how to argue AI safety with logic, evidence and live metrics.