How X’s “Under the Hood” Settings Feature Tackles Long-Standing Shadowban Rumors
Active creators can now download JSON files detailing exactly if and how visibility-limiting labels have been applied to their account or specific posts.
Summary
X expanded its open-source GitHub codebase by 10 to 15 times, releasing core ranking mechanics under an Apache v2 license.
A new settings feature allows active creators to download JSON files detailing any visibility-limiting account labels applied over the past month.
X is pulling back the curtain on its content distribution mechanics with a massive update to its open-source ranking algorithm. The platform expanded the codebase powering its default “For You” timeline and released the files under an Apache v2 license. This latest push increases the available code by 10 to 15 times compared to previous transparency efforts. Engineers and researchers can now access intricate details regarding model configuration, filters and the core ranking systems. These components dictate exactly which parameters weigh different signals to decide what content surfaces on a user screen.
Alongside the code dump, the company introduced an “Under the Hood” transparency tool designed to tackle long-standing rumors of shadowbanning. Users who post at least 10 times in a given month can navigate to their app settings and download a JSON file outlining aggregate statistics. This file explicitly reveals if any visibility-limiting labels have hit their account or specific posts over the past calendar month. Non-technical creators can feed this data into a large language model alongside the GitHub repository code to decode exactly how the recommendation engine treats their content.
X Vice President of Product Keith Coleman noted that the release provides the core ranking logic that pulls and ranks posts for any given timeline. External researchers have already utilized the release to train and run the Phoenix scoring system locally. While the platform invites developers to submit pull requests to improve the algorithm, certain proprietary safeguards remain locked away. Systems relying on Grok to predict rule violations are deliberately excluded to prevent bad actors from reverse-engineering moderation filters and flooding the feed with spam.
This unprecedented level of algorithmic transparency shifts the power dynamic between social media platforms and their user bases. By equipping creators with both the underlying recommendation code and personal account data, X transforms a traditionally opaque tech ecosystem into an inspectable environment. The pilot program for the transparency tool is currently restricted to established accounts over a year old but will roll out more broadly soon. The move coincides with the platform resuming public reporting of monthly active user metrics following its recent corporate integration with SpaceX.
