Proposal to document Twitter algorithm parameters

As everyone knows, Twitter has been aggressively policing lately for automation and whatever pisses them off, but it appears to be on something of a schedule. Any one person cant really quantify it on its own but if we all did a little reporting back, we might be able to accurately quantify what's happening and when. For example, I have been consistently following/unfollowing for several weeks without pushback but today got an account lock on my first follow pass. I'm assuming some of this is on a schedule as there were no behavioral differences. Any thoughts or ideas how we could maybe pull something together?

This will require lots of accurate testing
It will cost you money and time to find out, which by at that point I think one should keep it to himself and develop his own private bot

The way to go on would be to create a public Google Doc with the current findings and post that on BHW
People can help and OP can edit the doc with further findings

Start out with basic limiting like how many accounts per IP, follow/unfollow limits, delays,..
and then move on to successful behaviour which is tougher to automate

I don't do Twitter anymore, but when i did (a year ago), there was a well-recognizeable pattern in suspension waves. The waves hit in every other week on Tuesday or Wednesday, almost like clockwork. When it was over and if your accounts survived, you could be sure, that the accounts are safe for another two weeks.

While i see where you're coming from with this idea, it can be dangerous too, if you share something complex like this publicly. Their reps can easily see it and when you think, that you figured everything out, they switch things up and you're back to square one.
It's better, if you keep your findings to yourself or share it with people in private who you trust 100%.

It's occurred to me that publishing results widely could be problematic and potentially self defeating. I was thinking more along the lines of having a reporting thread here. Of course that doesnt prevent Twitter's corporate spies from seeing it but it would limit the audience. In the end, there is a rhyme and reason behind the policing though and there's enough smart folks like us who could fuck 'em back and god knows that would be worth the effort it seems. Fair points, all.

twitter users update status whenever they like to, so it will generate a flow of tweets. If we use batch algorithm to aiming at solving this problem, we propose an online algorithm, called online-LLDA, which incrementally updates model parameters when a document arrives.

THat would necessitate testable parameters that indicate the state ghosting etc. None of those parameters appear to be external

I used to be into Twitter big time and I've almost finished a twitter bot I'm working on so I'd be really interested in this. I found previously it was mostly follow churning, IP addresses and spam reports and DM frequency.

There's some very sharp people around here who could contribute and maybe help to turn some tables back in our favor, possibly. This thread is really an exploratory thing, seeing what people feel about it and offering thoughts about how this could be done. My thought had been to find a way to quantify things a little and look for trends in people's feedback, but perhaps there are ways to automate this which would be far better. Any and all thoughts and resources are welcome.