X Opens More of Its Ranking Algorithm and Adds Tool to Show Visibility Limits

X is opening up significantly more of the technology behind its For You feed and testing a new transparency feature designed to show users when labels applied to their accounts or posts may be limiting visibility.

The company released an expanded version of its recommendation system on GitHub on Thursday under the Apache 2.0 open-source license. The repository includes core ranking code, configuration parameters, filtering systems, and other infrastructure used to determine which posts appear in auser’ss For You timeline.

The release goes further than X’s previous open-source efforts by exposing some of the weights used to combine predicted user actions into a score for each post. X’s repository also includes code that determines whether a post is displayed, removed from the feed, or placed behind an interstitial.

X reveals more of how the For You feed works.

According to X’s documentation, posts in the For You feed come from both accounts a user follows and accounts they do not follow. The system gathers candidates from those sources, ranks them using its Phoenix model and then applies additional filtering before assembling the final timeline.

Phoenix evaluates a user’s recent activity and predicts the likelihood that the person will take different actions on individual posts. Those predictions are then combined using weighting parameters that X has now made available in the repository, providing researchers and developers with greater visibility into how the ranking process works.

The latest release also includes code used to train and run the Phoenix model, along with synthetic data that allows developers to perform proof-of-concept training outside X.

Under the Hood shows visibility-related labels

Alongside the code release, X is piloting a feature called Under the Hood that gives users aggregate information about labels applied to their accounts and posts that can affect visibility.

Eligible users can download information about those labels as a JSON file, enabling them to compare their account data with the systems documented in X’s public repository. The tool could provide more clarity for users who suspect their posts have been restricted or what is often described as being “shadowbanned.”

The pilot is initially limited to a test group of accounts that are at least one year old and have posted at least 10 times during the previous month. X plans to expand availability more broadly after the initial testing period.

The company also suggests that users who are not comfortable reading code can use an AI assistant to help interpret their report alongside the public repository.

Some ranking systems remain private.

The release does not expose every system that can influence distribution on X. The company’s GitHub documentation states that some moderation-related systems remain private because publishing them could make it easier for bad actors to manipulate the platform or evade enforcement.

X says the combination of public code and visibility reports is intended to provide greater insight into how posts are distributed without revealing information that could facilitate abuse.

Developers can also submit proposed changes to the open-source project on GitHub, allowing external contributors to suggest improvements to parts of the recommendation system.

The transparency push comes after years of debate over how social networks rank and distribute content. Questions about algorithmic amplification, political content, moderation and undisclosed reductions in visibility have followed both X and its predecessor, Twitter.

The expanded repository does not end those debates, but it gives researchers and users more material to examine. X says its goal is to make the systems governing post distribution easier to audit, critique and understand while continuing to protect parts of the platform that could be exploited if fully disclosed.