LoL BP / M4WORLD CHAMPIONSHIP · SEQUENTIAL DRAFT
From first ban to deciding game.
Choose teams and starters, explore next-action probabilities, then record each game winner to continue the series.
PATCH 26.20
Choose an edition to enter the draft
The pre-event edition pins this delivery; the updated edition uses the latest published snapshot. Both currently share the same inputs. No automatic live-match feed. Each series keeps its starting version.
Loading draft data…
Map side and draft order are independent in S16. Change them and substitute players between games. S15 confounds the two: this version models draft order; map side is recorded/displayed without a separately estimated effect.
Classic changes legality only. The action policy was trained on Full Fearless; Classic probabilities are unvalidated.
Team logos
Kept on this page only. PNG, JPEG or WebP, up to 2 MB.
Previously picked champions · unavailable in later games
LINEUP CHECK
Confirm this game’s roles
Drag champions to swap roles, or use the role menus. Confirm all five roles for both teams before recording the winner. Earlier action probabilities stay unchanged.
Draft complete. Who won this game?
You supply the winner. First to 2 wins in BO3 or 3 in BO5. Picks carry Fearless locks; bans do not.
Probabilities describe this next action, not winning or M1 tournament presence. Role shares estimate flex assignment. Contribution numbers are model scores, not percentage points. Auto-simulation conditions every later step on earlier simulated choices.
Opponent tendencies and seasonal styles
Head-to-head choices are compared with the same team against other opponents and shrunk for small samples. Positive stage index: earlier; negative: longer-game tendency. These are adjusted descriptive associations.
S15 evaluation and limits
Next-action audit on 28 knockout games. Heat calibration uses only earlier play-in/Swiss labels and excludes the training row’s entire series. Existing upstream versions already used S15 feedback, and these knockout games were inspected in the previous iteration, so this is a repeated retrospective conditional audit, not blind end-to-end validation. S16 is unvalidated.
Download errors and baselines