Video explanation
Posts become tokens traveling through tracks, scoring and filtering, ending with a rank and a served or cut result.
Candidates before ranking
Recommendation systems gather potentially relevant posts before ranking them for a viewer. The token pool and tracks represent these stages.
Multiple signals
Engagement, relationships and content can inform scoring. The model’s scales and weights are not current production X parameters.
Filtering after scoring
Rules and mixing can still act after scoring, so a rank does not guarantee that a post is displayed.