The Forever Open — Live
Swiss qualifier · best-of-1 duels · top 4 per class reach the Finals. Everything below updates on its own — no refresh needed.
Live now
Games being played right now, most-watched first, with the model's call. Bar = left player's chance.
Head-to-head predictor
Pick any two players: the model combines their spec matchup with each player's form so far.
Class vs class
Each cell = row class's win rate vs the column class.
Green = row favoured, red = unfavoured. Faded = fewer than 10 games. Overall excludes mirrors. Byes, no-shows, ties and tentative wins excluded.
Spec vs spec
Specs with 40+ games. Spec = talent tree with the most points. Hover a cell for the record.
Faded = fewer than 8 games.
Most lopsided spec matchups
The 15 most reliable lopsided matchups (picked by 95% lower confidence bound, 12+ games, so tiny samples can't sneak in), sorted by win %.
Most played specs
Qualifier standings
Per-class cut: record, then Buchholz (opponents' strength). Green rows are currently inside the top 4; dashed line = cut.
Finals bracket · projected
The official bracket is drawn after qualifiers end. Until then this builds it from the current top 4 per class, using the official rules: 36 players seeded by match points, then wins, then Buchholz; seeds 29–36 play a Play-In; then a 32-player double-elimination bracket. Odds come from the player model; title odds from simulating the whole bracket thousands of times (the Losers Bracket is approximated with random pairings).
Predictions & performance
A stats model, not a crystal ball. It scores every game against the player's spec matchup and how strong that specific opponent has proven to be, so a Warrior beating a 5-0 Demo Warlock earns far more credit than a Warlock beating a 0-4 Warrior.
How the numbers work
Model: P(A beats B) = spec-vs-spec matchup + A's skill − B's skill (log-odds). Spec-vs-spec records are thin, so each is blended with the class-vs-class rate. Every player's skill is fitted jointly over all games (a Bradley–Terry rating), with a prior pulling it toward "average for their spec" until they've played enough games.
Expected W = what an average player of the same spec would have won against these exact opponents. +/− = actual wins minus that: positive means outperforming your spec and your schedule.
Win chance = estimated chance to beat an average opponent (spec strength + personal skill). Projected wins = current wins + remaining Swiss rounds × win chance (ignores that Swiss pairs winners vs winners, so leaders run a little high). Rankings need 3+ decided games.
Title favourites · highest win chance
Biggest outperformers · beating their class's odds
Projected finalists · top 4 per class by projected wins
Unluckiest so far · furthest below their class's odds
Upset alerts
Wins the model gave the winner 35% or less going in (that game backed out of both ratings). New upsets also pop up in the corner while you watch.
Latest upsets
Biggest upsets so far
Win rate by race
Raw win rate, plus the rate the race's players should have from their class matchups. The gap is the race's effect with class stripped out (Undead is mostly Paladin/Warlock, so its raw rate flatters it).
Talents & builds
Level-35 builds (26 points) from every checked-in player. Colour = share of that class's players with at least 1 point in the talent.