AIBRIEF
The GitHub repo is 10 to 15 times the size of X's 2023 code drop. A new in app page tells users whether their account was hit by 'ranking labels' (the platform's own term for shadowbans), but not which signal pushed them down.
X put the code that runs its algorithmic "For You" feed on GitHub on Thursday under an Apache v2 license. The codebase is X's largest open-source release yet, roughly 10 to 15 times the size of its 2023 recommendation code drop, per TechCrunch. The repo includes the core ranker, model configuration, filters, and the signal-weighting parameters that decide which posts surface in the feed.
Alongside the code, X added an "Under the Hood" page in app settings that tells a user whether any ranking labels were applied to their account or posts in the past calendar month. Users who posted ten or more times in that window can download a JSON file of aggregate stats. X suggests non-technical users hand the file to a large language model pointed at the GitHub repo for interpretation.
External researchers familiar with recommendation systems had already run the core score outside X before the launch. VP of Product Keith Coleman told TechCrunch he expected some reaction: "This is this kind of thing that I think people will be fairly shocked that we are releasing."
The audit surface is asymmetric. Researchers can read the ranker, but a normal user can only see whether a label was applied, not which signal pushed their score down. The transparency tool is also a pilot, limited initially to accounts at least a year old, and the JSON export is gated to active posters.