I noticed that quite a few accounts (sample of about 25 accounts) have had blocks and ev/pv actions taken against them after about 12-24hrs after stopping all actions. As in, the action was taken after the accounts were already on rest for quite a while.
I’m thinking that actions may be taken against accounts after a random delay, to make it more difficult to know which action caused the block.
E.g. If you have 10 accounts simultaneously automating with similar behavior/limits. They all triggered the algo somehow and are flagged to have {insert random annoyance/block/verification here} implemented. Then a random delay of x is given for the punishment to be implemented, while even allowing some accounts to continue freely, this would make it very difficult to link the punishments to specific actions.
I tested this delay with a small sample of ~4 accounts, not warming them up and just following 200 people on day 1, one account made it all the way while others with identical settings got blocked at different amounts of followers between 9 to ~80.
Proxy definitely seems to be a large factor. IMO it really seems like there is some randomness mixed in though and I suspect it is to prevent us from understanding what the tripwires are.
Alternatively they might be split testing detection methods and seeing which ones have the least false positives, probably calculating it from how many ‘report issue’ submissions they get on the action blocks.
Anyhow, just some speculation 
What do you think?