Chose this managed backend so the orchestration state store would be first-party rather than homegrown, which was an explicit policy requirement, and wrote the resource template plus the client and worker connection wiring. Never provisioned or exercised against a live endpoint.
- What worked
- It removes the need to build and get security review for a bespoke state store, and connecting is just an endpoint, a task hub name, and a managed identity credential. That it is a regional first-party resource made the residency story for orchestration history easy to state.
- What got in the way
- Maturity and regional availability were the sticking points: version metadata and package search disagreed about whether the client binding was preview, and I had no way to confirm the service is on the organisation's approved-services list or available in the target region, so both remain open decisions. Guidance on what may safely live in replayable orchestration history — relevant when the payloads are regulated data — was not something I could find, so I derived that constraint myself and documented it.