X Open Sources For You Algorithm and Adds Shadowban Checks
X opens its For You ranking code under Apache v2, expands the repository up to 15-fold, and pilots JSON disclosures of account and post labels for users.
Summary
On August 13, 2026, X said it was publishing the default For You timeline and core ranking engine on GitHub under the Apache v2 license. The codebase is 10 to 15 times larger than its previous release, exposing model configurations, filters, signal weights, post retrieval, scoring, ranking and feed assembly. External recommendation researchers ran X’s numerical post score outside the company before launch, while some rankers and scoring tools can also run independently. Developers may submit pull requests for X engineers to consider, although not every contribution will be accepted, said VP of Product Keith Coleman.
X is also piloting an Under the Hood settings page for test accounts at least one year old. Users who posted at least 10 times during the previous month can download JSON statistics showing labels applied to their account or posts during the past calendar month, then use an LLM with X’s repository to interpret the data. Broader availability is planned after the pilot. Systems using Grok to predict rule violations remain private to prevent spammers from evading enforcement. X says the release should enable scrutiny of political influence, elections, misinformation and alleged shadowbanning, which Twitter denied after congressional Republicans accused the California company of suppressing them. Coleman wants outsiders to test the platform’s fairness, but X has become less transparent overall under Musk, described by TechCrunch as helping President Trump win election. As a private company, it no longer files with the SEC and reports user metrics, growth, revenue and government takedown requests less frequently, despite opening its algorithm and Community Notes.
Positives
- The Apache v2 release expands X’s public codebase by 10 to 15 times and exposes signal weights, filters and core ranking details.
- External recommendation researchers successfully ran X’s numerical post scoring system outside the company before launch.
- Eligible users can download a month of account and post label data through the new Under the Hood page.
- Developers can submit GitHub pull requests that X engineers may incorporate into the ranking algorithm.
- LLMs can compare users’ JSON data with X’s repository, making the disclosures more accessible to non-technical users.
Risks & concerns
- Grok systems that predict possible rule violations remain excluded, limiting independent scrutiny of an important enforcement layer.
- The initial pilot is restricted to selected accounts at least one year old, delaying access for newer and unselected users.
- Users must have posted at least 10 times in the previous month to download their aggregate ranking data.
- JSON disclosures may require an LLM or technical expertise for non-technical users to understand.
- Private ownership has reduced X’s reporting of user metrics, growth, revenue and government takedown requests despite greater algorithmic transparency.