According to Beating, X's official algorithm repository released a second major commit on May 15, four months after Elon Musk's January pledge to open-source the platform's recommendation algorithm. The update spans 187 files with 18,263 new lines and 926 deletions, advancing from architectural documentation to functional inference pipelines and ad insertion logic.
The Phoenix recommendation model now features an end-to-end demonstration pipeline that performs candidate recall, ranks content by predicting engagement metrics (likes, replies, retweets, dwell time), and computes final ranking scores. Notably, the update fulfills Musk's original commitment by adding ad-mixing code showing that ad placement is governed by safety intervals, adjacent content risk, brand safety rules, and keyword policies—not fixed positions. Additional modules for spam detection, post classification, and policy enforcement were also introduced.