The split-horizon point deserves emphasis — it is the failure mode that bites hardest, because the same name answering differently inside and outside means an incident can look like an application bug while it is really a resolution difference, and no single console shows both views at once. Two things I would add to the patterns section: first, treat TTLs as part of the design, not a default — a short TTL on records that move between clouds limits how long a stale answer can linger after a failover, while long TTLs on stable names keep resolver load down. Second, test the hub-and-spoke forwarding path with the same rigour as the data path: forwarders are a single point of failure for every cross-cloud lookup, so run two resolver endpoints in different zones, health-check them, and alert on NXDOMAIN spikes and forwarding latency, which are usually the first sign that a zone or a forwarder has drifted. IaC is the right call for keeping the zones in sync — without it, the drift you describe is a matter of when, not if.