infrastructure CPU calculations don't take into account the processes running on the nodes. Instead of the current calculation, find all workload CPU/memory usage not in the openshift-* namespaces, then subtract that from the cluster CPU usage to get infrastructure use.
Metrics workload:cpu_usage_cores:sum workload:memory_usage_bytes:sum are added openshift:cpu_usage_cores:sum openshift:memory_usage_bytes:sum now include all CPU/memory on nodes not explicitly part of a workload.