X Open-Sources Expanded 'For You' Ranking Algorithm and Adds User Transparency Tool

X announced on August 13, 2026 that it is open-sourcing its 'For You' ranking algorithm and adding a transparency tool that lets users see if their account or posts have been impacted by its ranking systems TechCrunch.

The company published the source code for its 'For You' timeline on GitHub at github.com/xai-org/x-algorithm under the Apache v2 license TechCrunch. The expanded codebase adds model configuration, filter, and core ranking system details, including the parameters used to weight different signals, making it roughly 10 to 15 times larger than before.

VP of Product Keith Coleman said the release includes core ranking code that pulls and ranks posts for any given user and assembles the feed, and some systems like the ranker and score can be run outside the company TechCrunch. Ahead of launch, X previewed its open source codebase to external researchers familiar with recommendation systems, who were able to get the score (numerical value calculated for every post) running outside of X.

From the GitHub repository, developers can submit pull requests that X engineers will consider incorporating into its algorithm TechCrunch.

The open-source release excludes some systems, such as those that use Grok to predict whether a post could be violating a rule, to prevent bad actors from working around the rules to flood the network with spam TechCrunch.

On the user-facing side, X is rolling out a new transparency tool on an 'Under the Hood' page in the app's settings. Users who have posted ten or more times over the past month can download their aggregate stats as a JSON file showing any labels applied to their account or posts over the past calendar month TechCrunch. The tool will initially be available as a pilot to a test group of accounts at least a year old before rolling out more broadly.

X's Help Center page 'Debunking X myths' states that X does not shadow ban and that it ranks posts to create a more relevant experience, with users always able to see posts from people they follow X Help Center.

Researchers have previously said the recommendation algorithm published by X does not offer the kind of transparency that would actually be useful Engadget. That assessment, published in February 2026, preceded the current expanded release.

The question now is whether this substantially larger codebase, combined with the JSON-based account-label transparency tool, changes that assessment. In this author's view, the inclusion of weighting parameters and the ability for external researchers to run the ranker and score outside X's infrastructure do move the needle. Whether it moves it far enough to satisfy critics who want full visibility into content classification and moderation pipelines is a separate matter. The deliberate exclusion of Grok-based rule-violation prediction systems signals a boundary: X is willing to expose ranking mechanics but not the enforcement layer that feeds into them.

For developers and researchers who work with recommendation systems, the Apache v2 release and the pull-request workflow open a relatively unusual channel. Platforms of X's scale rarely invite external code contributions to core ranking pipelines. The practical impact will depend on whether X engineers genuinely merge community pull requests or treat the repository as a read-only transparency artifact. The GitHub repository itself, dated unknown, serves as the canonical source for the code, while the announcement and feature details come from the August 13 reporting TechCrunch.

The JSON download requirement, ten posts per month, sets a floor for participation that excludes casual or new accounts from the pilot. Combined with the one-year account age threshold, this narrows the initial transparency tool to a relatively established user base. Whether that is a deliberate choice to reduce noise or a simple scaling constraint is not stated.

What this enables, at minimum, is a structured way for technically inclined users to inspect the labels applied to their own presence on the platform. For the broader research community, it provides a more detailed view of X's ranking signal weights than was previously available. Whether that constitutes useful transparency remains contested, and the February researcher assessments will likely be revisited as the expanded codebase is examined in the coming weeks.