X published the current source code and ranking weights for its For You feed on August 13, 2026. On the same day, it began a limited test of Under the Hood, a report designed to show users account and post labels that may restrict visibility. Read together, the two releases make one point more concrete: feed distribution is not only an engagement prediction. Candidate posts come from different sources, receive scores, and then pass through a separate visibility-filtering layer.
The release does not give brands a guaranteed formula for reach. X's open-source repository explicitly says some rules are withheld to prevent people from gaming the system. Live experiments, model versions, and viewer-specific context also cannot be reduced to one code snapshot. The useful news is not a trick such as using a particular word to gain reach. It is a better way to diagnose whether a distribution problem is about ranking, filtering, or content-market fit.
What exactly did X publish?
The repository under the xAI-Org account brings together core parts of the For You feed. The updated package includes in-network and out-of-network candidate sources, ranking weights, visibility-filtering code, model training, a SimClusters similarity system, and architectural notes. The repository marks August 13, 2026 as its latest update.
The pipeline can be understood in three broad stages. It first retrieves candidate posts from accounts a person follows and from discovery sources outside that network. A model then predicts likes, replies, reposts, and other possible outcomes. The system applies different weights to those signals and combines them into a ranking score. Finally, safety, author visibility, user preferences, and other policy rules can allow a candidate, place it behind an interstitial, or drop it.
The published notes describe a boost for new-author content, a discount for out-of-network candidates, and separate weights for different engagement types. These should not be treated as fixed universal numbers that calculate live reach. The code is useful for understanding system structure. It does not mean every experiment flag and production setting is documented at the same time.
This Fark Studio illustration shows in-network and out-of-network candidates moving through ranking and visibility layers into a personal feed. It is not an X interface, user data, or an exact algorithm diagram.
Ranking and visibility filtering are not the same thing
This is the distinction brand teams most often miss. A low ranking score means a post falls behind other candidates. A visibility filter is a separate layer deciding whether a post or account can be shown in a particular context. In the code published by X, outcomes broadly include allowing content, showing it behind an interstitial, or dropping it.
Filtering is not limited to a platform-wide safety label. Accounts blocked or muted by the viewer, country and age restrictions, the relationship with an author, post state, and some integrity signals can also affect the outcome. The same piece of content therefore does not have to follow the same distribution path for two people.
The Under the Hood report is intended to expose part of that layer. According to X's announcement and the GitHub notes, the tool aggregates labels applied to an account and its posts that may limit visibility. Testing begins with a randomized, limited group of eligible accounts that are at least one year old. Meeting those conditions does not guarantee immediate access.
The visual abstracts how the same content candidate can end in an allowed, interstitial, or dropped state. It is not an official moderation interface or a real account decision.
What is confirmed and what remains uncertain?
X has clearly released more detail on its ranking and visibility-filtering components than its previous summaries provided. The code confirms that predicted engagements are converted into a combined score and that other controls operate after ranking. Under the Hood is intended to report labels that could affect visibility to the account owner.
The uncertain area is wider. We do not know whether a particular weight applies identically to every user on every date, how extensive the unpublished anti-abuse rules are, or when the limited report will expand. Seeing a label also does not prove it is the only cause of an audience decline. Topic interest, network structure, creative quality, timing, competition, and negative user feedback can change the outcome in the same period.
Reducing open code to a content-hack checklist would therefore be a mistake. Artificially maximizing an engagement type with a visible weight may create negative feedback and integrity signals over time. The more durable value for a brand is separating distribution and filtering hypotheses, then testing them with documented experiments.
What should brands in Türkiye do now?
1. Preserve three to six months of posts with date, format, topic, author, in-network and out-of-network reach, engagement, and qualified outcomes. Do not rely only on total impressions. 2. If Under the Hood is available, archive a dated capture of the report and its label explanations. Compare label changes with the reach timeline for affected posts. 3. Do not treat a visibility-filter possibility as a ranking problem. If there is a warning, account state, or content restriction, check the policy and appeal route first. If there is no label, do not automatically call weak reach a hidden penalty. 4. Avoid changing topic, format, and publishing time in the same test. Small experiments with one controlled variable produce stronger learning than algorithm rumours. 5. Do not make likes or reposts the final objective. Measure site visits, qualified enquiries, video completion, branded search, or meaningful community responses separately. 6. Reduce dependence on one powerful account. Build original ownership across expert authors, employee advocacy, and the brand account rather than copying the same post across a network. 7. Add negative feedback, mutes, unfollows, and reports to creative-quality review. High short-term engagement does not guarantee sustainable distribution. 8. Monitor the repository as a monthly change log, but do not treat every commit as a strategy change. Act when an official announcement, live account data, and repeatable results align.
This audit framework combines a content archive, code changes, dated experiments, and human review in one loop. It is not a performance dashboard published by X.
Where should teams wait?
Under the Hood is a limited test. Do not make it a mandatory audit step until availability broadens and teams understand which labels appear and how quickly they update. For accounts without access, third-party shadowban tests should not be treated as official diagnosis.
Teams should also wait before changing creative formats in bulk based on weights in the repository. The live system includes experiments and unpublished rules. There is no reason to wait before improving measurement archives, separating reach from qualified outcomes, and preserving policy records.
Fark Studio perspective
The window X has opened is useful when it moves social publishing from an algorithm guessing game to a measurable content system. A social media team can own creative hypotheses and community feedback, content production can protect originality and format quality, and digital marketing can connect distribution to site and conversion outcomes in the same experiment record.
If you want to separate X visibility from algorithm rumours and examine it with account evidence, plan a social distribution audit with Fark Studio. The objective is not to chase one weight. It is to decide in advance what evidence should trigger which response when reach changes.
Sources
X Open Source, For You feed transparency announcement, August 13, 2026; open-source update and the limited Under the Hood test.
xAI-Org GitHub, X Algorithm, updated August 13, 2026; candidate sources, ranking weights, visibility filtering, and architecture notes.
xAI-Org GitHub, Under the Hood label transparency tool, August 13, 2026; eligible accounts, randomized test group, and report scope.
Social Media Today, X Shares New Insights Into Transparency and “Shadowbanning”, August 13, 2026; independent summary for brands and creators.
