Draft the stronger team composition.
Draftsmith is a draft engine for League of Legends tournament play. It knows all ten players, scores both teams' compositions as a whole (lanes, synergies, how the two line-ups match up) and searches for the bans and picks that take away the opponent's answers and build the biggest advantage in team composition, given the ten players in the game. It shows ranked options with its reasons; the team decides.
Prototype.
What Draftsmith does
A draft is a two-team game of twenty moves, played before the game itself. Draftsmith treats it as one: it plans its own side and the opponent's at the same time.
Scores both compositions
Champion strength on the current patch, lane matchups and synergies, judged for both teams together.
Knows all ten players
Champion pools, mastery and recent ranked games of every player, so a pick is valued for the person who would play it.
Plans against the best reply
Finds the bans and picks that take away the opponent's best answers to your composition.
Ranks options, with reasons
Each suggestion shows the win chance after the move and why it scores well. The team makes the call.
How it works
Learned models judge a draft position; a search uses them to find the moves that hold up against the opponent's best answers. Each part does one job.
Draft so far
Bans, picks and whose turn it is
Ten players
Pools, mastery, recent ranked games
Patch data
About 230k games
Composition model
Strength, matchups, synergies, team profiles
Player model
How well each player plays each champion
Opponent model
Their best answers and likely bans
Search over goal compositions
Both sides plan; moves that remove the opponent's answers score higher
Self-play drafter
Reinforcement learning, in development
Ranked options
Win chance after each move, with reasons
The team decides
Draftsmith advises; it never locks in
- Composition model trained on about 230k games
- Scores the ten picks together: per-patch champion strength, lane matchups, synergies and team profiles.
- Player model
- Champion pools, mastery and recent ranked history, so the same champion is worth more in the hands of its main.
- Opponent model
- Plans against the best answers the opponent has, not their habits, and forecasts the bans they are likely to make.
- Search
- Looks for goal compositions for both sides at once, and for the moves that steer toward yours and away from theirs.
- Self-play drafter in development
- A reinforcement-learning drafter trained by playing drafts against itself, to think further ahead.
- Light model on this site
- A player-blind version of the composition model that runs in your browser, for the practice and live pages.
Twenty decisions in sixteen turns
Tournament draft has twenty bans and picks in a fixed order. Four times a team makes two of them back to back (R1 + R2, B2 + B3, R3 with its fourth ban, B4 + B5), so Draftsmith plans those as one move: sixteen turns in all.
Read the engine's reasoning
Three drafts the prototype engine played against itself, on anonymous example teams. For each step: its options, the win chance after each, and the goal compositions both sides are steering toward. Test yourself: guess the engine's move.