X tests shadowban transparency tool as it open-sources ranking algorithm
The social media giant is providing users with a new 'under the hood' settings page to check for visibility-limiting labels, coinciding with a broader push to open-source its recommendation systems.
X is testing a new feature that allows users to verify whether their posts have been subject to reach restrictions, commonly referred to as shadowbans. The tool is currently being rolled out to a small group of testers via an 'under the hood' page within the app's settings, offering aggregate statistics on visibility-limiting labels applied to accounts and posts over the preceding month.
The feature enables users to download their data as a .json file, which details how the platform's algorithm has treated their content. While the data is not provided on a per-post basis, it reveals specific labels that have impacted visibility. For instance, users can identify if their content has been flagged with NSFW, spam, violence, impersonation, or hateful conduct labels, which may restrict access to non-followers or underage users.
This rollout coincides with X open-sourcing additional aspects of the ranking algorithm that powers the 'for you' timeline. The company’s vice-president of product, Keith Coleman, described the move as providing an unprecedented level of transparency into the X algorithm. The open-sourced code includes labels for NSFW content, spam, violence, impersonation, hateful conduct, and a 'civic integrity' flag for posts violating policy after user reports.
The 'civic integrity' label is reserved for critical escalations and restricts a post's visibility to the author's profile only. It is applied to posts detected in response to user reports as containing content in violation of X's Civic Integrity policy. The code also lists specialized labels, such as one for emergency incident response, though no specific labels for political content appear in the current release.
Despite the move towards openness, X continues to withhold certain information that could be used to manipulate the algorithm. The underlying code for advertisements and other systems outside the 'for you' timeline has not been released. Elon Musk has previously stated that every line of code touching the X system would be open-sourced and third-party audited by August, a goal the current release does not fully meet.
X plans to invite third-party recommender system experts to preview and pressure-test the code release to ensure clarity. However, it remains unclear how these experts will be selected or if their findings will be made public. The company notes that previous open-source releases of its recommendation algorithm offered limited insight into its actual operations.
