What was published
The release in brief
It is published in the xai-org/x-algorithm repository on GitHub.
The core of For You is readable
What was published is the core code that determines which posts a viewer sees in the For You feed. It combines posts from accounts the viewer follows with out-of-network posts discovered through machine-learning retrieval, filters them, and then a transformer model (the mainstream AI architecture today) ranks them. Ranking and the question of whether a post may be shown at all are separate paths; the latter is handled by visibility filtering.
"This repository contains the core code that determines which posts a viewer sees in the For You feed on X. It combines in-network content (from accounts the viewer follows) with out-of-network content (discovered through ML-based retrieval and other mechanisms), filters content based on a variety of inputs, and ranks posts using a transformer model."(opening)／"Ranking sets the order. Whether a post can be shown at all is decided separately, by `visibility-filtering/` …"(Overview section)／"Licensed under the Apache License 2.0."(License section) — from the X For You Feed Algorithm README
What the August 13 update added
The August 13 update added key configuration parameters, including the weights used to blend predicted action values into a score, along with code for the systems that affect whether a post is filtered. Alongside that, Phoenix — which handles candidate retrieval and ranking — was replaced with the training code for the models the feed actually uses, and synthetic data generation code was included so a proof-of-concept training run can be done locally.
"Adds key configuration parameters (including weights used to blend predicted action values into a score for a post)"／"Adds code for systems that impact whether a post is filtered from the For You feed"／"Replaces the Phoenix demonstration model with the code used to train the models the feed uses, as well as synthetic data generation code so one can run a proof-of-concept training run of Phoenix."(August 13th, 2026 section) — from the X For You Feed Algorithm README
How to read the weights
A common misreading, and the reality
The company added a clarification the next day, in response to a widespread misunderstanding.
Weights scale predicted probabilities, not counts
The documentation states plainly that the weights scale predicted probabilities and do not scale raw engagement counts. So reading it as "a report has 468 times the weight of a like, therefore one report cancels out 468 likes" is incorrect. Predicted probabilities depend on your own behavior, so the same post can work differently for different people.
"The weights scale the predicted probabilities of such actions (or predicted continuous values, e.g. dwell time) — they do not scale the raw engagement counts, so e.g. it'd be incorrect to see that a report has 468 times higher weight than a like and conclude that e.g. \"1 report cancels out 468 likes\". The weights are a multiple on your own predicted probability of Liking, Reporting, etc, which is substantially driven by your own behavior."(August 14th, 2026 section) — from the X For You Feed Algorithm README
The filters are readable too
In accordance with Brazilian electoral law, For You runs a filter that removes posts from accounts reported to Brazil's Electoral Court for the 2026 election, unless the viewer explicitly follows the account. X frames this as a benefit of open source: you can see that changes like this exist and exactly how they work.
"in accordance with Brazilian electoral law, For You now runs `Brazil2026ElectionFilter`, which removes posts from accounts reported to Brazil's Electoral Court for the 2026 election, unless the viewer explicitly follows the account. A benefit of open-source is that you can see that changes like this exist, and exactly how they work"(August 14th, 2026 section) — from the X For You Feed Algorithm README
What is withheld, and how to check your own account
Where the line is drawn
Not everything is out in the open.
What is not published
Publishing code that affects distribution invites people to try to outmaneuver it. To reduce that risk, the specific prompt templates used in Grox (X's internal AI processing) and some of the rules in the botmaker rule engine are not published.
"One challenge with making code that impacts post distribution public is that people could use it to try to game the system. To reduce the risk of this, there are a limited set of files not currently published in the repository, e.g.:"／"Grox prompts. E.g. the j2 files with the specific LLM prompts used in Grox."／"Some botmaker rules"(What's not in this repo? section) — from the X For You Feed Algorithm README
Checking your labels in Under the Hood
"Under the Hood" opens at x.com/i/under_the_hood and shows aggregate statistics about the visibility-impacting labels on your account and posts. What people commonly call a shadowban can be checked against official aggregates rather than guessed at. It is a pilot, and availability is stated to expand over time. Once you know your labels, you can match them against the published code and follow the effect.
"We're piloting a new transparency tool that lets people see aggregate statistics about the visibility-impacting labels on their account and posts."／"The tool is available here — we'll be shaping it based on your feedback and expanding availability over time."(Under the Hood Label Transparency Tool section; the "here" link in the original points to https://x.com/i/under_the_hood )／"… one can match any labels present on their account to the code to understand if or how the visibility of their posts is affected, and critique it if desired"(What's not in this repo? section) — from the X For You Feed Algorithm README
The README is long, and reading it end to end is work. To convert it into markdown with its structure intact, the following tool can help.
