Each rating receives a score as a function of its distance to the item’s reputation-weighted mean rating (computed after its inclusion). The score increases as the distance decreases, with quadratic growth near the mean.

A rater’s reputation is defined as the Wilson lower bound of the confidence-weighted average of their scores.

Each rating can be (a) reassessed if the item’s mean changes, updating its contribution to reputation; or (b) restarted, reapplying it with current reputation, affecting the item mean but not the rater’s reputation.