During investigation for BIZ-654, I discovered that ROSA HCP clusters appear to be half-sized when counted via cluster:usage:workload:capacity_physical_cpu_hours. cluster:usage:workload:capacity_virtual_cpu_hours values look fine though.

Note I used cluster:node_instance_type_count:sum{_id='$uuid'} to lookup node counts and then searched the web to find number of vcpus for each node, and used this to get an "expected" value, and observed ^.