X open sources its ranking algorithm and gives users a tool to see applied labels
X published the ranking engine behind its For You feed on GitHub under the Apache v2 license, and the codebase grew to 10 to 15 times its previous size. At the same time users gained a tool showing whether labels have been applied to their account or posts. The long running shadowban debate can finally be held against data.
The codebase grew 10 to 15 times
Per a TechCrunch report dated August 13, 2026, X released its For You feed algorithm and core ranking engine on GitHub under the Apache v2 license.
The published codebase is 10 to 15 times the size of the previous version. It includes model configuration, filtering systems and the weights of the different signals that determine which posts appear in a feed.
Per X VP of Product Keith Coleman, users can see the systems that filter out potentially problematic, rule violating content, and developers can even run them outside the company.
How the label tool works
The feature added to the app Under the Hood settings lets users who posted at least 10 times in the past month download their aggregate stats as a JSON file.
That file reveals whether any labels were applied to the account or its posts during the past calendar month. Per the report, non technical users can upload this data to a language model alongside the X GitHub repository to have it interpreted.
The direct target of the tool is the long running shadowban accusations. Users can now see concrete evidence of how their visibility was affected, seeing which labels or restrictions were applied rather than suspecting invisible suppression.
Transparency has a limit too. Per the report, systems using Grok to predict rule violations remain private, to stop bad actors from circumventing moderation.
Transparency will not reward good content, but it ends the guessing
The most concrete benefit of this move is that it changes the ground of the debate. Until now, when reach dropped, the first explanation on the brand and creator side was almost always a shadowban, and there was no way to verify it. Now there is data to look at. That assessment is ours.
Second, the practical limit of open sourcing. Publishing the code does not change how many people can read and understand it. Per the report non technical users are pointed toward a language model, so the transparency tool needs another tool in order to be interpreted.
Third, the Grok based systems staying closed. The part people most want to know, why a piece of content was deemed rule violating, remains invisible. Transparency is partial.
Fourth, the real gain on the brand side is diagnosis. Being able to tell whether a campaign lost reach because of content quality, timing or an applied label affects budget decisions directly. That comment is ours.
What it means for businesses in Türkiye
The assessment below is not in the sources, it is our reading. The report contains no Türkiye specific information.
X still sets the agenda here and a significant share of brands speak organically on it. When reach drops suddenly the interpretation is usually intuitive. This tool at least shows definitively whether a label was applied.
The practical suggestion: if your brand account posted at least 10 times in the past month, download this data from settings and establish a baseline. If there is no label, the cause of a reach drop has to be sought in content and timing. If there is one, examining which type of post triggered it becomes meaningful.
A reminder for corporate accounts: because the rule violation prediction systems stay closed, you will not see why a label was applied. The tool diagnoses, it does not explain. Expectations should be set accordingly.
It is also useful to read this alongside X ending revenue sharing in favor of original content. The platform is redefining both its payment rules and its visibility rules in the same period. That connection is our comment.
The UNALSOFT take
On the brand management side, the flaw we see most often in social media reports is reach drops being explained without evidence. The sentence the algorithm changed explains everything and says nothing, because it cannot be measured. This tool makes at least one variable measurable: was a label applied to your account or not? If the answer is no, the remaining causes are about content, and content can be worked on. We use data like this to reduce the number of guessing sentences in our reports, because a reason that cannot be measured does not produce a strategy.
What is behind your drop in reach?
Let us read your social data together and stop guessing.