How do scores and feedback change a recommendation ranking?

A recommendation ranking is an ordered list produced by a specified rule. Scores need a meaning within that rule; they are not automatically probabilities of enjoyment or measures of an item’s quality. This transparent toy uses three fictional items and two visible features, topic match and freshness.

Initially the score is two times match plus freshness. A has features three and one, giving seven points. B has two and three, also giving seven. C has one and two, giving four. Letter order breaks the tie, so A precedes B, followed by C. The first two items are displayed.

The optional feedback is one recorded click on A or B. The toy adds a two-point bonus to that item for the next ranking. A click on B changes its score to nine while A stays at seven and C at four. B then moves ahead of A. This is a fully stated arithmetic update, not an inferred model of every person’s preferences.

Ranking affects exposure: only displayed items can receive an observed click in this example. C’s lack of feedback therefore cannot be read as evidence that the viewer disliked C. Treating exposure-shaped observations as neutral preference data can create a feedback loop, where earlier choices influence the data used for later choices.

Real recommendation systems use many different objectives and methods. The lesson isolates scoring, tie-breaking and one feedback event so their relationships can be inspected. It does not claim that the toy formula describes a real service, guarantees useful recommendations or resolves the broader problem of biased feedback.