X Opens Its Ranking Algorithm and Adds a Shadowban Checker

What X Is Releasing and Why It Matters

X, the company formerly known as Twitter, is making a much larger portion of its recommendation machinery public. The code behind the default "For You" feed, including the core ranking engine, is being posted on GitHub under the Apache v2 license. The repository now includes model configuration details, filters, and the parameters used to weight different signals when deciding which posts appear for each user. X says this expands the open-source codebase by roughly 10 to 15 times compared with its earlier efforts.

Alongside the code, X is rolling out a transparency tool in the app's settings under a page called "Under the Hood." Users who have posted at least 10 times in the past month can download an aggregate statistics file in JSON format showing whether any labels have been applied to their account or posts during the previous calendar month. The feature is initially being offered to a pilot group of accounts that are at least a year old. X suggests that non-technical users can paste the JSON into a large language model, point the AI at X's GitHub repository, and ask for an interpretation.

X's vice president of product, Keith Coleman, said external researchers were able to train and run the company's Phoenix scoring system using the open-source code before the launch. However, the company clarified that those researchers did not receive per-post scores. Some systems are deliberately excluded from the release, including those that use Grok to predict whether a post may violate rules, which X says protects against bad actors gaming the moderation system.

What Open-Sourcing X's Ranking Code Actually Changes

Why X Is Making the Ranking Code Public Now

The move lands against a long history of accusations that the platform manipulates which accounts and posts become visible. Before Elon Musk bought the company, Republicans in Congress accused Twitter of "shadowbanning" their content, while the company repeatedly denied doing so. In the Musk era, concerns have continued over how the algorithm influences politics, elections, and misinformation. Publishing the weighting parameters gives outside researchers a concrete way to audit whether certain accounts or viewpoints are being systematically deprioritized. VP of Product Keith Coleman frames the release as an attempt to make the system "audited" and "critiqued," with the goal of building trust.

What the Release Does and Does Not Cover

The newly public repository lets developers inspect how posts are filtered and ranked, and X says engineers will consider accepting pull requests that improve the algorithm. This could lower the cost of refining the feed by drawing on outside expertise. But the release is not a complete look inside X's moderation stack. Systems that use Grok to predict rule violations remain private, and the company has not shared the per-post scores generated by its Phoenix system. That means independent reviewers can examine the ranking logic, but cannot verify every content decision or reconstruct the exact visibility score X assigned to a specific post.

Who Gains and Who Still Faces Limits

Users who suspect their reach has been restricted gain a concrete diagnostic tool: if labels have been applied to their account or posts, they can see those labels rather than relying on anecdotal evidence. Developers and researchers gain a meaningful sandbox for studying feed mechanics, and X gains a potential stream of external code contributions and a credibility argument in political debates. The main losers are those expecting full algorithmic transparency, because the withheld Grok-based moderation components and per-post scores leave important questions unanswered. X's history of being less forthcoming with user metrics and government-takedown reports since going private also means that skeptics may see this as a selective transparency campaign.

What Users, Developers and Researchers Can Do With X's New Tools

• If your account is at least a year old and you posted 10 or more times in the past month, check the app's settings for the "Under the Hood" pilot. Download the JSON file and use a large language model pointed at X's GitHub repository to interpret any labels applied to your account or posts.
• Developers can review the Apache v2-licensed ranking code and submit pull requests. X says its engineers will consider incorporating improvements, but the Grok-based systems for predicting rule violations are deliberately excluded.
• Treat the new transparency tool as partial rather than complete. Because the Grok moderation components are not public, you cannot use this release to verify every visibility or content-removal decision.
• Researchers planning to work with the Phoenix scoring system should note X's clarification that external researchers trained and ran the system without receiving per-post scores.
• For everyone else, the practical next step is to monitor whether the pilot expands beyond the initial one-year-old test group and whether label data becomes available for a longer history than the past calendar month.