Kubernetes pod scheduling drift after node autoscale events

After upgrading to K8s 1.31, we're seeing pods get scheduled to newly provisioned nodes but then rescheduled within 30-60 seconds. Looks like the autoscaler's taint/untaint cycle is interfering with the scheduler's scoring phase. Topolvm for local storage, Cluster API for node provisioning. Anyone hit this? Considering adding a pod disruption budget or tweaking the descheduler's eviction policy. What worked in your cluster?