Run the step forever under the policy: a completed step resets the backoff and reruns at once, since the step owns its own pacing. tryAny leaves asynchronous exceptions alone, so cancellation tears the loop down.

One loop's supervision policy. A loop classifies a fault that no retry can fix, such as an unconfigured handle reached at runtime, as Permanent, and everything else as Transient.

The policy for a loop with no wiring fault to fail up on: every synchronous escape is residue, logged and retried at schedule's pace.

What the supervisor does with a synchronous fault the step let escape. An asynchronous exception is never classified, so cancellation propagates and the shutdown race always wins.

A bounded exponential backoff, doubling from the base towards the cap as consecutive failures mount, so a persistently-failing dependency retries at most once per cap interval.

The delay before the next retry, given how many failures ran consecutively: base * 2^failures, saturated at the cap.

The pace a background loop retries a transient fault at: one second after the first failure, doubling to a thirty-second ceiling.

A delay list as a Control.Retry policy: retry n waits the n-th delay in microseconds, so the list's length is the retry budget. It paces a bounded run, not an endless loop.