X For You Feed Algorithm

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.

Table of Contents

• Notable Updates
• October 26th, 2026
• October 13th, 2026
• October 8th, 2026
• September 18th, 2026
• August 14th, 2026
• August 13th, 2026
• Overview
• System Architecture
• Request Path
• Labeling Path
• Components
• How It Works
• Scoring and Ranking
• Filtering
• Experiments and Configuration
• What's not in this repo?
• Under the Hood Label Transparency Tool
• Key Design Decisions
• License

Notable Updates

October 26th, 2026

• Brazil 2026 Elections filter removed. With the 2026 Brazilian elections over, we have removed Brazil2026ElectionFilter from For You. Posts from the accounts it covered are no longer filtered out because of it. See the August 14th, 2026 update for what the filter did.

October 13th, 2026

• Retrieval trained to surface posts people read all the way through. Phoenix retrieval is now also trained to surface posts that people read all the way through.

October 8th, 2026

• Ranking predicts time spent much better. The Phoenix ranking model now does a better job of estimating how long you'll spend on each post: see the code.

September 18th, 2026

• Under the Hood. Reports now include information about whether one's account or posts have had their visibility limited because of required compliance with law(s). For example, you'll be able to see if any of your posts were withheld from showing in a country following a legal demand — and which country.

August 14th, 2026

Notable updates:

• How weights work. There's a common misconception about how weights related to actions (e.g. Like, Share, Block, Report, etc) work in ranking. 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. We've added comments to the code so that LLMs or people reading it are more likely to understand it correctly.
• Brazil 2026 Elections. As announced by X, in accordance with Brazilian electoral law, For You began running Brazil2026ElectionFilter, which removed posts from accounts reported to Brazil's Electoral Court for the 2026 election, unless the viewer explicitly follows the account. (Removed after the election; see the October 26th, 2026 update.)

August 13th, 2026

This release:

• 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.

Among new systems included are:

1. Visibility filtering: visibility-filtering/ determines whether to show a post, drop it, or show it behind an interstitial.
2. The systems that produce labels that drive visibility filtering's responses: rules that apply labels (botmaker/, botmaker-rules/, scarecrow/), models that score accounts on various dimensions (agatha/, bdsm/, user-cred-v2/), models that examine images and video (media-model-proxy/, clip/), and enforcement (abuse-enforcement-service/).
3. Phoenix model code: phoenix/ now contains code that trains and runs the model, plus synthetic data generation.
4. SimClusters: simclusters/, an additional source of posts from accounts the viewer does not follow that is called in retrieval alongside Thunder and Phoenix retrieval.

This update is also paired with a new Under the Hood transparency tool that allows people to see aggregate statistics about the labels on their account and posts that can limit visibility.

Overview

The For You feed is assembled per request. Posts come from two places:

1. In-Network — thunder/ keeps recent posts from the accounts a viewer follows in memory
2. Out-of-Network — phoenix/ retrieval and simclusters/ find posts from accounts the viewer does not follow. popular_posts_source adds a shared list of popular recent posts.

Both are ranked together by the same model. Phoenix reads the viewer's recent engagement history and predicts, for each post, how likely the viewer is to take each action on it. Those predictions are combined into one score using weights held in the code — see Scoring and Ranking.

Two pipelines do the work. The Post Pipeline finds, ranks and filters posts. The Blending Pipeline wraps it and adds what the model does not rank: ads, Who to Follow recommendations, prompts.

Ranking sets the order. Whether a post can be shown at all is decided separately, by visibility-filtering/, from the viewer's own actions such as blocks and mutes and from labels that other systems here attach to posts and accounts.

System Architecture

Request Path

┌──────────────────────────────────────────────────────────────────────────────────────────┐
│                                   FOR YOU FEED REQUEST                                   │
└──────────────────────────────────────────────────────────────────────────────────────────┘
                                             ▼
┌──────────────────────────────── HOME MIXER   home-mixer/ ────────────────────────────────┐
│                                                                                          │
├───────────────────────  POST PIPELINE   PhoenixCandidatePipeline  ───────────────────────┤
│                                                                                          │
│  ┌────────────────────────────────────────────────────────────────────────────────────┐  │
│  │ 1. QUERY HYDRATION                                                                 │  │
│  │    user action sequence — the viewer's recent engagements, and the                 │  │
│  │    main input to the model · following list · blocks and mutes · muted             │  │
│  │    keywords · posts already seen and served · followed topics, etc.                │  │
│  └────────────────────────────────────────────────────────────────────────────────────┘  │
│                                            ▼                                             │
│  ┌────────────────────────────────────────────────────────────────────────────────────┐  │
│  │ 2. CANDIDATE SOURCES — queried in parallel                                         │  │
│  │    ┌───────────────────────────────┐ ┌────────────────────────────────────────┐    │  │
│  │    │ IN-NETWORK                    │ │ OUT-OF-NETWORK                         │    │  │
│  │    │ Thunder                       │ │ Phoenix retrieval   retrieval model    │    │  │
│  │    │   recent posts from the       │ │ SimClusters         cluster similarity │    │  │
│  │    │   accounts the viewer follows │ │ Popular posts       shared list        │    │  │
│  │    └───────────────────────────────┘ └────────────────────────────────────────┘    │  │
│  └────────────────────────────────────────────────────────────────────────────────────┘  │
│                                            ▼                                             │
│  ┌────────────────────────────────────────────────────────────────────────────────────┐  │
│  │ 3.