Two designs, and the zone that is not the zone
A partner consumes an internal service over PrivateLink from their own account. To keep latency down and avoid cross-zone charges, both teams agreed to put everything in one zone and wrote it into the standards document: us-east-1a, both sides, no exceptions.
The endpoint works. Latency is measurably worse than the single-zone design predicted, the cross-zone line on the bill did not go away, and on one occasion a newly created endpoint could not see the service at all.
Boundaries, outermost first: Consumer account: Partner app (us-east-1a), Interface endpoint (us-east-1a) Provider account: Endpoint service (us-east-1a), Quote service (one zone) Connections: Partner app calls Interface endpoint Interface endpoint publishes to Endpoint service — one zone each Endpoint service calls Quote service
Boundaries, outermost first: Consumer account: Partner app (us-east-1a), Interface endpoint (use1-az4) Provider account: Endpoint service (us-east-1c), Quote service (one zone) Connections: Partner app calls Interface endpoint Interface endpoint publishes to Endpoint service — same physical zone Endpoint service calls Quote service