A broker can be alive while one of its replicas is still behind the leader. Being reachable is different from being in sync. The ISR is a set of replicas currently meeting the synchronization requirement, and it changes during failures and catch-up. A returning broker must recover its assigned logs before those replicas count toward the minimum ISR for durable writes. These tasks begin after a broker restart but before replication catch-up. Use the local replication control and inspect the resulting ISR. The control completes recovery immediately; it does not model bandwidth, fetch lag or recovery throttles. An operational check should verify partition health rather than treating a green process as complete recovery.
Inspect the topics, partition logs and group offsets before changing the cluster.