Not sure exactly what you mean, but I run a setup where everything is mostly containers, except for HA which runs in a VM as a native install. I would prefer to run other services that connect to it separately (in their own Linux VM+Docker).
It's probably not a discrete number or even heuristic (like number of years you've been in good standing). Rather it's probably another part of their machine learning backend that tracks trends and user behavior as a whole.
But as a guess, if you're a new account, maybe 2 or 3 times. For every year you've been a member, probably 1.5 times that number per year up to some max (let's say, 7).
Standard retaliatory tactic so they can tell investors the ex-employees are actually the bad apples and of course not the company.
The impression I get is that the only secret ML actually has is that they are years away from showing anything to validate their worth (by which point many competitors will have shown their cards and removed the 'magic' from the leap).