X's Recommendation Algorithm, Broken Down Into Three Machines

X's recommendation algorithm is often described as a black box. Ryan posted a breakdown video on X that splits the pipeline into three parts: Thunder, Phoenix, and Simcluster. Three diagrams first connect retrieval, scoring, and filtering.

Ryan explains the three algorithms very plainly. Thunder is responsible for pulling posts from followers, starting with the circle of people you know. Phoenix looks at what strangers might like and is key to whether a hit can break out beyond your existing audience. Simcluster groups things by topic and vertical, explaining an old question: why does distribution become more stable when you stay within one niche topic?

The video also has two more parts: The Judge and The Bouncer. The Judge is the scorer: actions like likes, replies, and shares add points, while scrolling past, muting, and blocking subtract points. The Bouncer handles filtering; content outside the 48-hour window gets blocked. According to Ryan, once you understand this system, viral hits can be reverse-engineered.

The tactical advice is direct too. Post about topics people are willing to discuss and share; get to the point quickly at the start; and don't pile up posts in a short period, because your posts will compete with one another. In the comments, one user said the Simcluster part was what they most wished they had known earlier, and that staying on one niche topic helps reach more than posting tricks do. Another user said the biggest takeaway is that posts also compete with your own posts for traffic, so don't expect every post to perform equally well on the same day.

The algorithm is more like a scoring system. You can't control every stranger, but you can control whether your topics are focused and whether your opening is worth stopping for. If you want to study viral hits, first turn your account from a general store into a specialty counter.

Source: Ryan's X post