X Expands Open Source Codebase for For You Feed and Adds Account Ranking Transparency Tools
Social media network X announced on Thursday that it is significantly expanding its open source codebase by releasing the source code for its default “For You” feed algorithm and core ranking engine on GitHub under the Apache v2 license. Alongside the expanded code repository, the platform is introducing a transparency feature designed to allow users to see if ranking systems or labels have impacted their accounts or specific posts.
What Happened
X released a substantially expanded repository that is roughly 10 to 15 times larger than its prior open source releases. The newly published code includes model configurations, filters, the core ranking system, and specific parameters used to weight signals that determine which posts appear in feeds.
In an interview with TechCrunch, X’s VP of Product Keith Coleman stated that the release covers the core ranking code that gathers, ranks, and assembles posts for users, as well as the filtering systems used for potentially problematic or rule-violating content. According to Coleman, some components, such as the ranker and scorer, can be run externally.
In addition to publishing the code, X is rolling out an “Under the Hood” page in the application settings. This pilot tool will enable eligible users who have posted at least 10 times in the past month to download their aggregate account statistics in a JSON file format. The file reveals whether any moderation labels were applied to their accounts or content during the preceding calendar month.
Key Highlights
- Larger Codebase: The open source release under the Apache v2 license on GitHub is approximately 10 to 15 times larger than previous releases and includes signal weighting parameters and model configurations.
- Account Status Insights: An “Under the Hood” settings tool allows qualifying users to download a JSON file detailing any labels applied to their posts or accounts over the past month.
- Community Contributions: Developers can submit pull requests via GitHub, which X engineers will review for potential inclusion in the ranking algorithm.
- Exclusions for Safety: Systems utilizing Grok to predict potential rule violations have been withheld from the public release to prevent bad actors and spammers from circumventing platform rules.
- External Testing: External recommendation system researchers previewed the repository and successfully trained and executed X’s Phoenix scoring system using the open source code.
Why This Matters
The release is designed to address long-standing concerns regarding how algorithmic feeds manage the distribution of political content, elections, and misinformation. By providing visibility into ranking systems and parameter weights, the platform aims to allow the public to audit, critique, and evaluate content distribution fairness. Additionally, the downloadable account metrics offer users direct visibility into whether platform filters or labels have affected their reach, addressing past claims and allegations regarding hidden content suppression or shadowbanning.
What to Watch Next
The transparency tool in the “Under the Hood” section is initially launching as a pilot for a test group of accounts that are at least one year old before a broader rollout is implemented. X engineers will also begin reviewing pull requests submitted by developers on GitHub to determine if any external code contributions will be integrated into the live ranking architecture.
Frequently Asked Questions
What parts of the X algorithm have been open sourced?
The release on GitHub includes the “For You” feed source code, core ranking engine, model configurations, filters, and parameter weights used to assemble feeds, licensed under Apache v2.
What is the new “Under the Hood” feature?
It is a transparency setting that allows eligible users with at least 10 posts in the previous month to download a JSON file showing any moderation labels applied to their account or posts over the prior calendar month.
Are all of X’s systems included in the public repository?
No. Systems that use Grok to predict rule violations have been omitted to prevent spam and rule evasion.
Source: TechCrunch report and GitHub repository documentation.