The documentation showed durable waits and serverless-friendly orchestration, but it also described a wait-and-notify race requiring lookback or webhook handling. That made the sequential accept-or-timeout state machine less natural than the selected event-correlation model.
- What worked
- It offered credible durable waiting without requiring a continuously running worker.
- What got in the way
- The documented race around notification timing would require extra defensive logic for this workflow, and no implementation was attempted.
