← Back to seasons

How ratings work

Scores combine Kitsu user ratings with weighted votes and an adjustment for small samples. These scores differ from Kitsu’s average rating.

From ratings to votes

Each rating contributes one vote, split between upvotes and downvotes. Ratings use five bands. This weighting explains fractional vote counts.

Vote weights per Kitsu rating
Rating / 10UpvotesDownvotes
8.5–1010
6.5–80.750.25
4.5–60.50.5
2.5–40.250.75
0.5–201

Users Rated counts submitted ratings. Favourites counts Kitsu users who mark an anime as a favourite. Both show raw counts.

Laplace and Wilson scores

Laplace

Laplace scores add five to total votes before calculating a percentage.

100 × upvotes / (upvotes + downvotes + 5)

For example, 75 upvotes and 25 downvotes produce 71.43%. More votes reduce the effect of this adjustment.

Wilson

Older archives use the lower bound of the Wilson confidence interval with z = 1.96. This accounts for uncertainty when few ratings exist.

Season dashboards use their archived scoring method. Full history uses Laplace where recorded, otherwise Wilson. Historical scores stay unchanged.

Reading charts and rankings

  • Rating uses a 0–100% axis. Users Rated, Upvotes, Downvotes and Favourites share a separate count axis.
  • Graph toggles work independently from sorting. Rankings use each anime’s latest recorded value, then anime ID for ties.
  • Rating changes use percentage points: 70% to 72% is +2 pp. Count changes compare the previous recorded sample.
  • Season overview shows average ratings for new anime and leftovers separately. Zero scores are excluded from these averages.
  • Full history includes all archived seasons for an anime. Histories with more than 90 observations show each metric’s highest value per ISO week, Monday–Sunday UTC. Shorter histories show daily samples.
  • Weekly peaks can come from different observations. The 30/90-day views show daily samples. Gaps mark missing days or weeks.

Current and previous seasons update daily. Older seasons are archived and frozen. Downloads retain original observations, including values omitted from weekly charts.

Back to seasons