Well, if you battle-test the "what happens if unexpected reboot [edit: or network loss, or API failure]" side of things for EC2, and validate simulated resource failure for the other bits... wouldn't it basically amount to figuring out where AWS's *real* status dashboard is this month (eg, specific Twitter or HN thread) and mass-mailing realtime "there, there"s until everything mostly comes back up by itself?

Yes, in this position you'd best find a way to diversify some of them or move them to, say, a Proxmox instance.