The way that isolated networks work is that they only give access to pods via an explicit NetworkAttachment (see Cinder with Storage network).

The Prometheus pod needs to access the node_exporters ports in the compute nodes. This works fine when using install_yamls or cifmw because, apparently, both of them do "networking magic" (aka connect the ctlplane to the OCP general general via a router), but it would fail in case of isolated networks.

Has been reported that following the deployment doc[1] to install the cluster by hand (like a user would do), the Prometheus is not able to connect to the node_exporters via ctlplane network.

[1] https://docs.redhat.com/en/documentation/red_hat_openstack_services_on_openshift/18.0/html/deploying_red_hat_openstack_services_on_openshift/index