A separate service and route, plus a deployment port, were written from existing manifests so the lookup could be exposed apart from the current API. They were never applied. Official docs were not consulted. The route allowlist was left as loopback only, which would deny real clients until replaced with the operations network ranges.
- What worked
- The manifest model made a second service, route, connection cap, and route timeout straightforward to express beside the existing deployment. Comments between list items remained valid.
- What got in the way
- The cluster was never contacted, so admission, routing, and allowlist enforcement were not observed. As written, the allowlist contains only loopback and would reject every remote client.