Twitter's For You code tracked metrics to categorize users as “republican”, “democrat”, “power user”, or “elon” to keep algorithm tweaks from harming any group
Musk said he had no idea it was doing that! — Twitter has just released …
MashableMatt Binder
Context & Ripple Effects
Twitter’s partial release of its For You recommendation code made visible a set of group-level safeguards that were not apparent from the product itself. The release covered recommendation logic but excluded ad recommendations in the related coverage’s account of the partial code open-sourcing.
The finding also sharpens the limits of that transparency exercise: subsequent researchers said the released code lacked enough context to fully account for recommendations, as described in the critique of the code release’s missing context.
First-order effects
Twitter users and outside reviewers can see that For You evaluation tracked political and high-profile user categories when assessing whether algorithm changes would disproportionately affect a group.
Musk’s stated lack of awareness creates an immediate governance question for Twitter: who owns, documents, and approves sensitive segmentation embedded in ranking systems.
Second-order effects
The disclosure raises the standard for any platform presenting source-code publication as algorithmic transparency; code alone may reveal implementation choices without explaining the data, objectives, or operational controls behind them.
Political figures and power users gain a concrete basis to scrutinize whether ranking changes are evaluated differently across groups, increasing pressure on Twitter to explain its safeguards and definitions.
Third-order effects
If recommendation systems increasingly use protected or politically salient cohorts as guardrails, algorithm governance will move beyond publishing code toward auditable documentation of metrics, evaluation datasets, and decision rights.
The episode points to a durable tension in social ranking: safeguards meant to prevent group harm can themselves become a source of legitimacy risk when their categories and oversight are opaque.
The trend: This is part of the broader shift from demands for algorithm transparency toward demands for explainable, accountable governance of the metrics that steer ranking systems.
PS: People saying, “Meh, the module is just for collecting data on whether various A/B feature changes end up being good or bad for Elon personally” do not seem to understand how very entertaining it is to see Elon inserting “Benefit to Me” as an evaluation metric.
I asked Elon Musk why his name is hard coded here in Twitter Algorithm repo; you can listen to the answer yourself... and the Space finished right after that 🤭 https://github.com/... https://twitter.com/...
1. Likes, then retweets, then replies Here's the ranking parameters: • Each like gets a 30x boost • Each retweet a 20x • Each reply only 1x It's much more impactful to earn likes and retweets than replies. https://twitter.com/...
This has been a big priority for Musk — one he's specifically called out with individual scams that he's seen / others have pointed out to him. https://twitter.com/...
Like, he's just awful, but the thing is here, if he just said, “Yeah, I bought the fucking thing, it's mine, so I'll shove every one of my tweets into your timeline cos I fucking can”, I'd respect that. I'd block him (I have anyway) but I'd respect it. But this pretend nonsense🙄 …
Part of Twitter's algorithm specifically designates Jack Dorsey, Katy Perry, Stephen Curry and Barack Obama as “testing accounts” for getting random Tweets for testing, with an emphasis on Katy Perry in particular https://github.com/... https://twitter.com/...
Twitter didn't just open source their recommendation algorithm. They also revealed how you get a perfectly stable and fast CI: exit 0 https://github.com/...
elon was asked about this on Twitter Spaces and he said “this is the first time i'm seeing this” and said they should change it he also decided that the person who asked him about it would be the last speaker and ended the Space right after https://twitter.com/...
Unless the Democrat/Republican tags are assigned only to elected officials... wtf does this mean? Are the scrapping voter registration databases and cross referencing? Vibes? What? https://twitter.com/...
Told you 🤦♂️ This is because legacy checkmarks get a low weight and the system needed a way to differentiate Elon Musk since he has the same legacy checkmark ✅ https://twitter.com/...
It's even funnier to think they tried to remove all the places where Elon had been hard coded into the Twitter algorithm before releasing the code but missed one. $44B to become the most popular user on Twitter🤦🏾♂ ️https://twitter.com/... https://t.co/XdR8GhGesD