I suspect (but may be wrong, I don't know how trigger-happy the risk level changes are) the absolute number of events needed to trigger a risk level loss at the scale of Twilio would also represent a catastrophic number of arbitration cases.