I do wonder if we should do something more aggressive to encourage authors to assume that adapters are flakey, like specing them to start expired 10% of the time.

I also think it would be cleaner to change the one-requestDevice-success-per-adapter rule to one-requestDevice-attempt-per-adapter. I think it would be encourage more robust code if the workflow for requesting a device always required requesting an adapter.

In some ways, this turns an Adapter into a one-shot DeviceRequestConfigurator in a two-step device creation workflow.

Originally posted by @kdashg in #4352 (comment)