X has released a substantially larger version of the code behind its “For You” feed and begun testing account-level ranking disclosures, giving creators and researchers a closer look at how the platform distributes posts. The move, announced Thursday, Aug. 13, expands transparency around one of X’s most commercially important systems.
The public repository is published by xAI under the Apache 2.0 license and describes itself as the algorithm that powers the “For You” feed on X. Separately, a pilot feature called “Under the Hood” lets some users download data showing labels applied to their accounts and posts during the previous calendar month.
Creators can inspect more of the ranking machinery
TechCrunch reported that the code release is 10 to 15 times larger than X’s earlier public version and includes the core ranker for the “For You” timeline. The disclosure provides more detail on the systems used to retrieve, score and select posts for individual feeds.
The “Under the Hood” pilot is aimed at a practical question that source code alone cannot answer: what happened to a specific account. Eligible participants can review machine-readable JSON records containing labels that may affect distribution. The initial test is limited to accounts at least one year old that posted 10 or more times in the month, according to the report.
For creators, that distinction matters. A general ranking model can explain the rules of the road, but account-specific labels may reveal why reach changed after a moderation decision or detected behavior. Brands and agencies can use that information to separate content performance problems from platform enforcement.
The disclosure has a meaningful limit
The release does not include every component involved in ranking. TechCrunch reported that X is withholding a Grok-based predictor used to identify rule-violating content. That omission means outside researchers cannot fully reproduce or audit the system that determines when safety judgments affect recommendation.
Open-source code also does not automatically reveal how a live platform behaves. Training data, model weights, production settings, rapid experiments and internal policy choices can all change results. The public repository is therefore evidence about system design, not a complete audit of every user’s feed.
Transparency becomes a platform feature
Recommendation systems shape attention, advertising inventory and creator income, yet users usually see only the output. By pairing code with account-level disclosures, X is treating transparency as a product capability rather than a one-time publication.
The next test is whether the pilot reaches ordinary users and presents findings in a form that does not require technical expertise. JSON files may help researchers and sophisticated creators, but broad accountability requires plain-language explanations, appeals and clear links between platform rules and distribution effects.
For marketers, the immediate takeaway is to document audience shifts and review any disclosed labels before assuming that creative quality or media strategy caused a reach decline. For X, the larger challenge is proving that the released information is current, complete enough to be useful and matched by consistent enforcement.
Researchers will also need stable versioning and historical records. Without them, changes to ranking logic can be difficult to compare over time, weakening the value of a public repository as an accountability tool.