Explain the complete X recommendation algorithm pipeline. Use when users ask how posts are ranked, how the algorithm works, or want an overview of the recommendation system.
Use the skills CLI to install this skill with one command. Auto-detects all installed AI assistants.
Method 1 - skills CLI
npx skills i CloudAI-X/x-algo-skills/x-algo-pipelineMethod 2 - openskills (supports sync & update)
npx openskills install CloudAI-X/x-algo-skillsAuto-detects Claude Code, Cursor, Codex CLI, Gemini CLI, and more. One install, works everywhere.
Installation Path
Download and extract to one of the following locations:
No setup needed. Let our cloud agents run this skill for you.
Select Provider
Select Model
Best for coding tasks
No setup required
The X recommendation algorithm processes posts through an 8-stage pipeline to generate the "For You" feed. Each stage transforms, filters, or scores the candidate posts.
┌─────────────────────────────────────────────────────────────────────────────┐
│ X RECOMMENDATION PIPELINE │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ User Request │
│ │ │
│ ▼ │
│ ┌─────────────┐ │
│ │ 1. Query │ Hydrate user features, action history, socialgraph │
│ │ Hydration │ │
│ └──────┬──────┘ │
│ ▼ │
│ ┌─────────────┐ Thunder (in-network) + Phoenix (out-of-network) │
│ │ 2. Sources │ In-network: Posts from followed accounts │
│ │ │ Out-of-network: ML retrieval from all posts │
│ └──────┬──────┘ │
│ ▼ │
│ ┌─────────────┐ │
│ │ 3. Candidate│ Fetch tweet text, author data, visibility status │
│ │ Hydration │ │
│ └──────┬──────┘ │
│ ▼ │
│ ┌─────────────┐ │
│ │ 4. Pre-Score│ Age, duplicates, safety, blocked authors │
│ │ Filtering │ │
│ └──────┬──────┘ │
│ ▼ │
│ ┌─────────────┐ Phoenix ML → WeightedScorer → AuthorDiversity → OON │
│ │ 5. Scoring │ Each scorer adds/adjusts candidate.score │
│ │ │ │
│ └──────┬──────┘ │
│ ▼ │
│ ┌─────────────┐ │
│ │ 6. Selection│ TopKScoreSelector: Keep top N by final score │
│ │ │ │
│ └──────┬──────┘ │
│ ▼ │
│ ┌─────────────┐ │
│ │ 7. Post- │ Conversation dedup, previously seen, keywords │
│ │ Filtering │ │
│ └──────┬──────┘ │
│ ▼ │
│ ┌─────────────┐ │
│ │ 8. Side │ Logging, analytics, impression tracking │
│ │ Effects │ │
│ └──────┬──────┘ │
│ ▼ │
│ Feed Response │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Enriches the request with user context:
Two candidate sources provide posts:
// home-mixer/sources/thunder_source.rs
// Posts from accounts the user follows
served_type: Some(pb::ServedType::ForYouInNetwork)// home-mixer/sources/phoenix_source.rs
fn enable(&self, query: &ScoredPostsQuery) -> bool {
!query.in_network_only // Disabled for "Following" tab
}
served_type: Some(pb::ServedType::ForYouPhoenixRetrieval)Fetches full post data:
Removes ineligible candidates before expensive ML scoring:
AgeFilter - Too oldDropDuplicatesFilter - Duplicate IDsVFFilter - Safety violationsAuthorSocialgraphFilter - Blocked/muted authorsCoreDataHydrationFilter - Missing dataIneligibleSubscriptionFilter - Subscription required// home-mixer/scorers/phoenix_scorer.rs
// Calls Phoenix ML to predict engagement probabilitiesProduces phoenix_scores with 18 action probabilities.
// home-mixer/scorers/weighted_scorer.rs
// Combines probabilities into single score
weighted_score = Σ(weight × P(action))Produces weighted_score from action predictions.
// home-mixer/scorers/author_diversity_scorer.rs
// Penalizes multiple posts from same author
multiplier = (1 - floor) × decay^position + floorAdjusts scores to promote variety.
// home-mixer/scorers/oon_scorer.rs
// Adjusts out-of-network post scores
if !in_network: score *= OON_WEIGHT_FACTORBalances in-network vs out-of-network content.
// home-mixer/selectors/top_k_score_selector.rs
pub struct TopKScoreSelector;
impl Selector<ScoredPostsQuery, PostCandidate> for TopKScoreSelector {
fn score(&self, candidate: &PostCandidate) -> f64 {
candidate.score.unwrap_or
Keeps top K posts by final score.
Fine-grained filtering after selection:
DedupConversationFilter - One post per conversationRetweetDeduplicationFilter - One version per underlying postPreviouslySeenPostsFilter - Remove seen postsPreviouslyServedPostsFilter - Remove from current sessionMutedKeywordFilter - User keyword mutesSelfTweetFilter - Remove own postsNon-blocking operations after response:
Candidates start with:
├── tweet_id, author_id (from Sources)
├── tweet_text, metadata (from Hydration)
├── phoenix_scores (from PhoenixScorer)
├── weighted_score (from WeightedScorer)
├── score (from AuthorDiversity + OON)
└── Final ranking by score
pub struct PostCandidate {
pub tweet_id: i64,
pub author_id: u64,
pub tweet_text: String,
pub in_reply_to_tweet_id: Option<u64>,
pub retweeted_tweet_id
| Tab | Thunder (In-Network) | Phoenix (Out-of-Network) |
|---|---|---|
| For You | Enabled | Enabled |
| Following | Enabled | Disabled |
/x-algo-scoring - Detailed scoring formula/x-algo-filters - All filter implementations/x-algo-engagement - Action types and signals/x-algo-ml - Phoenix ML model architecture