If your feed feels like an endless loop of identical AI videos, that is not a coincidence. It is a feedback loop. Adjust the sliders below to see how small engagement signals snowball into a repetitive feed.
This is a deliberately simplified teaching model, not a replica of any real platform. Real recommender systems use thousands of signals, exploration mechanisms, and diversity rules. The core amplification idea, however, is real: what gets engagement gets shown more, which earns more engagement.
The source starts at five percent topic share with twelve percent engagement, sixty percent watch-time weight and twenty on its data-size slider. The daily multiplier is zero point one five one zero zero eight. The next share equals old share plus old share times one minus old share times that multiplier. Day zero, one and three shares round to fifty, fifty-seven and seventy-five out of one thousand illustrative slots. At four pixels per rounded slot the bars are two hundred, two hundred twenty-eight and three hundred pixels. These are calculated toy counts, not observed feed measurements. Across fourteen source updates, default share reaches about twenty-eight point one two percent, displayed as twenty-eight percent. With engagement set to zero and the five-percent start unchanged, share remains five percent. At five pixels per displayed percentage point, the comparison bars are one hundred forty pixels for twenty-eight and twenty-five for five. A zero starting share also remains zero because this recurrence has no external arrivals or exploration. These behaviors follow the fixed equation; no ranking model is trained or queried. The source clamps each updated share at ninety-nine percent. A one-hundred-percent initial share therefore becomes ninety-nine after the first step, even with zero engagement, despite the flat-share edge-message wording. Bars at five pixels per percent compare five hundred with four hundred ninety-five pixels. Four local quiz questions and the share URL teach the fixture; they do not prove a real feed forecast. Reset restores sliders but does not reset quiz answers. This is an authored simplified recurrence, not a replica of any platform, real user-data inference or model API.