Data 2026-11-12 10:52 UTC · pro games since 2024-10-01 ·
solo queue season 27 · meta: 208 tier-1/intl games since
2026-09-13 (patches 16.16, 16.17, 16.20)
How the scores work (0–100)
Comfort = 60% experience + 40% performance, for this player on this champion.
Experience = 100·(1 − e−n/6), n = pro games (weighted, half-life 1 year) + ¼ × solo queue games.
Performance = 50 + 250 × (win rate − the player's own win rate) + (GD15 − the player's own GD15)/40.
Win rates are shrunk to the player's baseline (6 pro / 15 solo queue games), so small samples stay near 50.
Meta = 70% presence rank (picks + bans in this role, percentile) + 30% win rate (40%→0, 60%→100, shrunk).
Comfort + Meta = 60% Comfort + 40% Meta. A champion the player never played scores Comfort ≈ 20.
Green ≥ 70, red < 45. Picks are saved in this browser only.
How the lane matchups work
Expected GD@15 = mean lane gold difference at 15 minutes for this champion pair in pro
games from all leagues since 2023-01-01 (Oracle's Elixir). Older games count less (half-life 180 days).
The table is symmetric: A vs B = −(B vs A).
Small samples are shrunk: each champion's lane average toward 0 (30 games), and each pair toward the
difference of the two champion averages (5 games). A pair never seen uses the champion averages.
Draft P(blue) = σ(logit(pre-draft P) + +0.051 × lane sum / 1000). The pre-draft P is lol-model's
map price when the match is on its card; you can type over it.
Backtest (lol-model out-of-sample map prices as the prior, 2026-07 to 2026-10 held out): the matchup term changes
log loss by about −0.0002 (±0.0002). The map price already holds almost all of it, so the shift is small on purpose.