Nice emphasis on pausing consumption rather than turning a scheduled SAP window into a DLQ cleanup exercise. One detail I’d validate is whether priority orders and bulk telemetry share Kafka partitions: pausing a shared partition can stall both, so separate topics or partitions (while preserving ordering where required) may matter as much as separate worker pools. I’d also treat the 80/20 split as an initial budget, then tune it against oldest-message age, consumer lag, SAP latency, and error rate. That gives operators a clearer signal for safe recovery than relying on a fixed ramp alone.