S16 Worlds · Player Champion Preferences
M3 · Player preferences and opposing bans anchored to the M1 meta
Project home · M1 champion meta · Method and reproduction
Predictions are conditional on playing; substitutes are not assumed to appear. Classic career data provides era-normalized preferences; direct training labels use Full Fearless only. Signature status or a high career win rate does not guarantee priority in a weak current meta.
Click a player portrait to inspect Worlds champion priorities.
Likely opponent bans when facing this team
Roster covers verified starters and currently confirmed public substitutes. Other club/academy players are not automatically registered for Worlds. Portraits come from official LoL Esports club pages.
Click a champion for career share, win rate, MVP, KDA, persistent preference, recent games and role first-pick. Availability differs by champion, so conditional pick probabilities need not sum to 100%.
S15 evaluation and actual pick-count comparison
Inspect actual pick counts by player
| Champion | M1 | M3 pre-event | Actual count | Forecast − actual |
|---|
Model and data coverage
A role-specific masked multinomial model predicts picks; an M1-anchored Bernoulli regression predicts opponent bans. Nine factor coefficients are shared across champions and players; there are no manual Faker/Azir scores. Chronological S15 validation selects a 45-day recent-preference half-life. The comfort-mixture coefficient is 7.66%, not an equal percentage-point bonus to each champion. Win, MVP and KDA use small-sample shrinkage. KDA is adjusted for role, league, patch, champion and the player’s year-wide performance, avoiding unqualified strong-team or zero-death bonuses. Repeated exact-role comfort retains an independent preference pathway with a mixture learned on S15.
MVP coverage is partial and currently limited to successfully matched LPL game awards; missing is neutral. Series MVPs are not assigned to one champion. Reputation uses dated sourced associations, has incomplete coverage, and may receive a learned zero weight. Available career archives begin in 2014; individual coverage varies and pre-2014 records are incomplete.
Forecasts average legal Full Fearless contexts from the past 90 days; this is not a specific-opponent, side-aware or BO5 game-one draft simulator. Champion-level bans combine the team’s five players and do not establish coaches’ targeting intent. Independent ban marginals do not yet force exactly five bans in a legal sequence. Recent professional practice and KR patch shifts enter through M1 without a duplicated correction.