Done when any of its policies is, and it stops asking at the first yes.

Mirrors tbp/monty/experiment/recognition_strategy.py::AnyPolicy. The result is the union of the answers it did get: a time_out_lms from an earlier policy survives a later one that says done on its own, so combining policies never loses a request.

Stopping at the first "done" is Monty's behaviour and it is observable — a policy after the one that fired is not consulted at all — which is why the recording compares the union rather than re-running every member.