[anigbrowl]

What tools do people use to administer these? Since Twitter and other platforms seem to have no interest in fixing the problem I figure I might as well stop trying to abide by their ToS and leverage the technology for my own agenda.

[kordless]

What needs to happen is that they open up all the blocked account information for all users, and show when users that are blocked by X many users have deleted tweets.

The problem defined is that Twitter allows viral communication to occur in Humans, but which is being exploited in a very specific way by a small group of people who understand how to spread dissonant ideas by intentionally using polarized, meme-based arguments, which themselves are viral in nature. It's like a logic "jingle" you get stuck in your head. Or a virus.

It's on Twitter to fix the Pandora's Box they have opened with this type of infrastructure. (And I'm still sticking to my claim that Twitter is infrastructure given how much it can govern our behavior.)

[anigbrowl]

OK, but any ideas on the tool front? I'm not being facetious, I've just decided that since it's a de facto free-for-all I might as well retaliate against bad actors rather than waiting for people to grow a conscience.

[dsacco]

If you're seriously asking: for the most part the tools are in-house. There isn't a single widely used framework (to my knowledge...) that covers everything. Basically, you'd use a mix of something like redis, beautifulsoup, digitalocean, ansible, a fleet of shady proxies and various middleware for everything. Throw in postgres too.

If you're not using the API, all you're doing is programmatically signing up for accounts, storing the corresponding login credentials, writing a little library that assigns all outgoing bot activities to a user-agent and proxy, queues their activities and executes them. Naturally you have a little library that manages the random profile information creator - collect CSVs from data.gov such that you can create convincing random names and addresses, then use random profile pictures from Google images.

The secret sauce is not in the orchestration (someone could wrap all this up into a framework pretty easily), it's in the structuring of activities so that you don't get caught by e.g. having your bots follow each other incestuously. That's a rather less automated process and require active vigilance and tweaking.

What I have seen work in the past is partitioning the botnet such that blocks of them slowly establish signal to noisy credibility in specific niches before intermingling. Many botnet creators attempt to create a multiplier effect, recursively improving their aggregate signal score by having the bots in a massive echo chamber with each other. This is easily caught.

Bot block A should have a few thousand from around the country commenting on the election and sometimes posting memes from reddit. Bot block B should retweet thinkpieces from the tech industry and be the first submitter of various obscure but passable Medium articles. And so on, and so forth. Basically, have the bots act like humans who mostly talk about one category of thing on Twitter, but who still have enough nuance to not seem spammy.

Once your botnet hits a critical mass, you no longer need to do this as strictly in the future. You can spin up new bot blocks quickly and have the mature blocks retweet and interact with them to bring them up to the requisite signal score more quickly. At this point you can capitalize on trends and have a credible mass of followers influencing a conversation on Twitter within hours or days of the trend emerging. As it snowballs, you compound the other blocks to simulate rippling popularity throughout the system (i.e. trends become less about one niche and more about everyone, like Uber->Uber's Sexual Harassment Scandal->Sexual Harassment).

The other secret sauce is in successfully managing the network on a budget, because while all those proxies are what diversify your botnet's origins, individual proxies typically score negatively for websites actively looking to reduce bots.