Why would you not just take a random sample, instead of cherry-picking scenarios that are guaranteed to be affected by selection bias?
So unfortunate.
That is some real BS gerrymandering on the part of Twitter.
Accounts that are not "monetizable daily active users" who spam and harass users on the service don't count?
Of course they count.
Just generate random numbers from 1 to 2^64 and use the "/users/show" API by user_id until you've got 100 successful hits that match a valid user, then once you have 100, score the users.
Do you get fewer than 5 out of 100 spam bots? That should be the test.
It could well be that 50% of accounts are bots, but as long as Twitter doesn't charge advertisers for them, they've been honest. But if their mDAU counts have been way off, advertisers will demand their money back.
If you want to get users to pay for the service, then you have to improve the product for the users, not for the advertisers.
And, to that end, the question of whether the service is being ruined by an abundance of bots and spam, regardless of whether it's negatively impacting _advertisers_, is a real issue if you're looking to acquire the business.