In both kube and spicedb writes, some errors don't make sense to retry (i.e. the request itself is invalid). We may want to do this as part of a larger effort to make error states clearer and more understandable.