How a Post Reaches Your Feed
The big picture: Post Pipeline + Blending Pipeline, two candidate pools
The Big Picture
Every time you open X, the For You feed is assembled fresh. The work is split between two pipelines:
Post Pipeline
Finds, ranks, and filters posts. This is where the ranking model lives.
Blending Pipeline
Wraps the Post Pipeline and adds what the model doesn't rank: ads, Who-to-Follow recommendations, and prompts.
Two Candidate Pools
Posts come from two sources, scored by the same model:
| Pool | Source | What it holds |
|---|---|---|
| In-Network | thunder/ | Recent posts from accounts you follow, kept in memory |
| Out-of-Network | phoenix/ + simclusters/ | Posts from accounts you don't follow, discovered via ML retrieval and cluster similarity |
Both pools are ranked together — there's no separate "following" ranking model. An out-of-network post can beat an in-network post if its predicted engagement is higher.
The key insight: the same scoring model evaluates every candidate, regardless of where it came from. The only difference is that out-of-network posts get a discount factor applied (see Lesson 4).
Check Your Understanding
1. What are the two candidate pools that feed the For You feed?
- A. In-network and out-of-network✓ correct
- B. Liked and retweeted
- C. Recent and trending
- D. Followed and suggested
Why: In-network (Thunder) holds recent posts from people you follow. Out-of-network (Phoenix/SimClusters) discovers posts from accounts you don't follow.
2. What does the Blending Pipeline add that the Post Pipeline doesn't rank?
- A. Replies and quotes
- B. Ads, Who-to-Follow, and prompts✓ correct
- C. Image filters
- D. Trending topics
Why: The Blending Pipeline wraps the Post Pipeline and inserts ads, Who-to-Follow recommendations, and prompts — things the ranking model doesn't score.
3. Are in-network and out-of-network posts scored by different models?
- A. Yes, each pool has its own model
- B. No, the same model scores both✓ correct
- C. Only replies use a different model
- D. It depends on the viewer's account age
Why: Both pools are ranked by the same model. Out-of-network posts get a discount factor, but the scoring logic is identical.
Exercise
Open your For You feed. Find one post from someone you follow and one from someone you don't. Notice how both appear in the same stream — the model ranked them together.
Hint: Look for the "Followed by X" or "You might like" labels that sometimes appear on out-of-network posts.