LONG THREAD: There's a simple algorithm Twitter could use to derank people with political stances they don't like. You might be aware of the recent "hack" (the "hacker" bought an admin account from its owner) that resulted in a few screenshots of Twitter's admin tools. 1/22

Unfortunately, Twitter seems to be removing screenshots of these admin tools, but of course you can search them up pretty easily and you can see that these admin tools can be used to effectively shadowban accounts. 2/22

But that's tedious. Can you imagine having to shadowban every single conservative account that gains popularity? Hundreds of accounts gain traction every second. Even with a large team, it would be impossible. But what if you could automatically derank wrongthinkers? 3/22

If you've ever been on Twitter, you know that one of its greatest features is the ability to reply to tweets. Someone spouts BS, someone replies calling out the BS, someone replies calling out their BS, etc. It's one of the best sites for crowd-sourced fact-checking. 4/22

But Twitter doesn't use the old forum method of ranking replies based on when the replies were sent. Rather, they have a ranking system that takes several factors into account. They keep a tight lid on the exact algorithm to prevent people from gaming it. 5/22

That said, we do know some of the variables that feature heavily in the ranking system. Obviously, likes, retweets, and replies factor in, but there are also account-based variables to try to reduce the damage bot-boosting services do to the rankings. 6/22

As a side note, Facebook also has these. Remember those “Russian bots”? It was a relatively small bot-boosting service with no ties to the Russian government that had no bias in clientelle. See the importance of crowd-sourced fact-checking on media stories? 7/22

But the main point is to show how a simple algorithm can allow a single Twitter employee to derank most conservative accounts and boost most left-wing accounts within a very short period of time, all by abusing the ranking system. 8/22

It starts by adding a variable to every person's account-based ranking, along with adjusting the existing algorithm to take the variable into account. If the variable is positive, the calculation is adjusted positively. If the variable is negative, it's adjusted negatively. 9/22

Simple so far. We're back to manually adjusting each account an admin likes or dislikes. To automatically adjust, understand that most engagements happen within a person's ideology group. Conservatives tend to like and retweet conservatives, for example. 10/22

All that is required now is to check for that variable every time someone likes or retweets, and then adjust each account's score based on that. The amount of processing required for this is very miniscule relative to the processing Twitter already does. 11/22

Here's a theoretical example: Donald Trump's score is equivalent to -10,000 likes. That is, he needs 10,000 more likes to be ranked equally with someone with a score of 0. Now every time you like or retweet one of his tweets, your score is adjusted as well. 12/22

For every tweet from Donald Trump that you have liked, you require 10 more likes (1/1,000th of the score) to be ranked the same as you were before. For every retweet, it's 100 (1/100th). If you like and retweet a single Trump tweet, your score rests at -110. 13/22

It gets even more efficient when you adjust both accounts, however. Every time Donald Trump likes a tweet, it's 100 off their score (the same as if they had retweeted him). Every time he retweets them, it's 1,000. One like and one retweet puts that account at -1,100. 14/22

Remember: This is for every (usually positive) interaction. Every like or retweet adjusts your score positively or negatively based on the other account's score. The more you like and retweet wrongthinkers, the worse your ranking. 15/22

And this is true on the positive end as well. If Barack Obama's score is 10,000, then the inverse happens. Every time you like an Obama tweet, your score increases by 10. Every time you retweet him, your score increases by 100. 16/22

If you create a new account and retweet 10 Trump tweets and 1 Obama tweet, your score will sit at -900. The next account you retweet will have their score adjusted by -90. The next account to like your tweet will have their score adjusted by -.9. 17/22

This is a simplified version. In all likelihood, the ranking adjustments would be logarithmic. That is, an account with a score of -10,000 would only be adjusted negatively twice as much as an account with a score of -1,000. 18/22

Manually adjusting even just a dozen accounts positively and a dozen negatively by very small amounts would very quickly result in the vast majority of people that are ideologically aligned with the manually-adjusted accounts having their tweet rankings affected. 19/22

This would not just affect replies, either. How likely a tweet is to show up in your feed is affected by its ranking. A heavily deranked account would have their tweets disappear from the majority of feeds, whereas a heavily ranked account would show up constantly. 20/22

Can this be proven? Unfortunately, Twitter doesn't allow anyone to see the ranking algorithm. But Twitter does let users see replies. And—for now—see various engagements in those replies. It's not hard to take that information and come to conclusions. 21/22

On Twitter, it's common to see contentious threads with unusual rankings in the replies. Tweets with a dozen replies, a hundred retweets, and a thousand likes will regularly outrank replies with 10 times as many of each. Do you see the pattern in the ideologies espoused? 22/22

@mitchellvii @ScottAdamsSays @MattBatzel @BasedPoland @NotReallyABear2 @TendiesB @Nyaranyar @KoreanAncap @I_am_onthisrock
Most probably won't see it, but figure it's worth a shot.