MyAnimeList
Unofficial ranking analysis
Ranking methodology explorer
Top 10
Method ranking
Comparison
Top title by method
Methodology
How the alternate rankings are computed
Uses the published score and rank from the saved MyAnimeList top-200 snapshot. This is the baseline for every rank-change and score-change comparison.
- Sort key: published MAL score.
- Vote shape is not adjusted.
- Best for matching the original source order.
Removes equal vote mass from both tails before recomputing the mean. The dashboard includes 10% each-side and 5% each-side variants.
- Low-tail votes are removed from 1 upward.
- High-tail votes are removed from 10 downward.
- Useful for reducing both hype spikes and protest votes.
Removes only the lowest 10% of votes, then recomputes the average from the remaining distribution.
- Tests sensitivity to low-score bombing.
- Does not reduce 10-vote fandom concentration.
- Scores are upward-biased by design, so use as a diagnostic view.
Combines tail dampening, vote-count stabilization, and a polarization penalty. This is the default methodology.
- Winsorizes the lowest 5% and highest 5% instead of deleting those votes.
- Shrinks toward dataset prior mean 8.655579 with prior weight k = 50,000.
- Subtracts 0.15 × winsorized standard deviation.
Robust score formula
robust_score = bayesian_score - (0.15 × winsorized_stddev)
bayesian_score = (n / (n + 50000)) × winsorized_mean + (50000 / (n + 50000)) × 8.655579
Higher vote counts rely more on the title's own distribution. Lower vote counts are pulled more toward the prior, which helps reduce volatility from narrow or self-selected audiences.