- Data window
- Preview sample — a one-time snapshot of 65,406 ranked games (Master+, EUW/KR/NA, patch 26.18), counted by GamGem while our Riot production key is under review. Figures will update daily once it is approved. Counted 26–27 Sept 2026; patches 26.18 (35,477) · 26.19 (18,580) · 26.17 (11,048) · 26.16 (293) · 26.15 (8); by server EUW 22,382 · KR 21,922 · NA 21,065 · EUNE 20 · TR 17. 642 games without a result — remakes ended by vote in their first minutes, and games recorded with no winner — were not counted.
- Rank aggregation
- Master, Grandmaster and Challenger, pooled — the three ladders the count read its players from. Splitting further leaves samples too small to separate a real effect from noise.
- Role assignment
- A champion is filed under every lane that holds at least a tenth of its counted games — the position the match itself reports, not the champion’s class. Its headline win rate is across all of them; each lane’s own win rate and games are counted too, and a lane’s tier list ranks by them.
- Counters
- The two champions in the same position on opposite teams, over that pairing’s own counted games — ranked on the edge pulled toward zero by its sample size, printed raw.
- Minimum sample
- None — no figure is hidden for being thin. Every rate prints its games, and the confidence band is its 95% margin: Very High within ±1.5 points, High ±2.5, Moderate ±3.1 (1,000 games), Low wider.
- Why a build is not a cause
- An item is counted when the match ENDS with it, so late items show long games and an item bought when already ahead shows games that were being won. Correlation is what a rate can show; it is not evidence that buying the item caused the win.
- Where the numbers come from
- Names, classes, abilities, items and runes are Riot Data Dragon. Win, pick and ban rates, builds, keystones, spells and matchups are counted by GamGem from Riot's ranked match records. An estimate, not a count: this panel is not in the preview sample, and is labelled so.