A good number of our flaky tests have a shared characteristic: the actual events recorded match the expected events, except the first event is missing.  My theory is that somewhere in the event listener setup we drop an event.