> With regard the "how to escalate?" issue: The company still struggles with corner cases like this. When you're building systems that cater for billion+ users, you have to look at systematic ways to deal with incoming requests such that the set of requests that need human intervention is as small as possible (assuming you don't want a company with millions of employees).

Maybe part of the lesson here is that it's simply impossible to scale to a billion+ users.