Interactions are batched in, a feed is requested per viewer, candidates are ranked by interest, and a visibility rule serves only what is published, processed and not deleted. A view attributes back to the session that produced it.
Exhibit 07 · Circo
The visibility rule serves only items that are published, processed and not deleted, so the gate holds this one back.
Interactions wait at the hopper. Choose an interest, whether a candidate is ineligible, and whether this continues the open session.
Interactions from across the platform are batched into the recommendation engine together, not sent one at a time.
A feed is requested for this viewer, carrying an interest in comedy. It continues the open session, so the next page follows on from the last.
The ranking drum orders the library by fit to comedy (short clips that make people laugh), the strongest matches first.
The visibility rule serves only items that are published, processed and not deleted. Every candidate here qualifies.
The surviving candidates drop into the feed tray in rank order, ready for the viewer to scroll.
A follow-up page continues the same session, so a view is attributed back to the recommendation that produced it.
Interactions from across the platform arrive at the hopper, batched together rather than sent one at a time. A feed is requested for one viewer, carrying an interest. The ranking drum orders the content library by fit to that interest, and the visibility gate applies a single rule: serve only items that are published, processed and not deleted.
The survivors drop into the feed tray in rank order. A dashed thread runs from the tray back to the session marker, because a follow-up page continues the same session, so a view can be attributed to the recommendation that produced it.
The model shows how feeds are shaped and attributed, not any real viewer's history.