The Worlds page gives every team a chance of each Swiss record and of reaching the knockouts, the final and the trophy. It then turns those odds into answers for the Pick’Em and the Crystal Ball. All of it comes from one model, rebuilt whenever new results arrive. This guide explains how it works and what it can’t see. Pick’Em and Crystal Ball are Riot’s prediction games in the League client, and StatChecked is not affiliated with Riot.
The data
The model reads every game of this season’s top-level leagues and international events from Leaguepedia’s public scoreboards. Two lower-stakes events count for less: the LPL’s preseason Demacia Cup at half weight, and the cross-region Demacia Cup Global Invitational at three quarters. During Worlds the season data refreshes every six hours and the event’s own results every ten minutes.
Game-by-game Elo
Every team has a rating, and the gap between two ratings becomes a win chance for a single game. A 100-point edge means about 64%, and a 200-point edge about 76%. After each game the winner takes points from the loser, more when the result was a surprise. In domestic games the step size is 20, so a 64% favourite that wins gains about 7 points, and one that loses drops about 13.
We rate games, not series. For a 60% favourite, a 3–0 win adds roughly 24 points. A 3–2 win leaves it about where it started, because it won exactly the share of games it was expected to. Games are processed in date order, so a team that improved during the season carries its later form into Worlds.
Region offsets
Domestic leagues are mostly closed worlds. LCK teams play LCK teams, so their results say how they compare with each other, not with LEC teams. To bridge the gap, each team’s rating has two parts: its region’s offset and its own share.
- Domestic games move only the teams’ own shares.
- Games between regions at international events, such as First Stand, MSI and the Esports World Cup, move the teams by a larger step (26) and also shift both regions’ offsets.
Say a Korean team expected to win 75% of games beats a North American team. The Korean team gains 6.5 points, the Korean offset rises by 1.5 and North America’s falls by 1.5, which nudges every team in both leagues. An upset moves the offsets three times as far. This is why international events matter so much to the model: they are the only evidence of how one region compares with another.
Regions start the year from priors based on recent Worlds: Korea +70, China +45, EMEA −10, North America −45, Asia-Pacific −50 and Brazil −80. A 150-point gap, such as Korea’s starting edge over Brazil, is worth about a 70% chance per game before either team has played.
Simulating the real format
With ratings in hand, the model plays the whole tournament:
- Play-In. Four teams play a double-elimination bracket, every series best of five. The winner takes the last Swiss seat.
- Swiss stage. Sixteen teams. In round one, pool 1 teams meet pool 4 and pool 2 meets pool 3, with no two teams from the same region. Later rounds draw opponents at random from teams on the same record, without rematches. Matches are best of one unless they can send a team through or out (when either team is on two wins or two losses), and then they are best of three. Three wins advance; three losses eliminate.
- Knockouts. Eight teams play best-of-five series, with the 3–0 teams drawn against 3–2 teams in the quarter-finals.
Each series is played one game at a time using the per-game chance, which also records how many games every team plays. The model runs the event 20,000 times and counts the outcomes. A team that reaches the knockouts in 12,400 runs has a 62% chance. With 20,000 runs, the counting error on any figure is under one percentage point.
Short series are where favourites stumble. A team with a 60% chance per game wins a best-of-one 60% of the time, a best-of-three 65% and a best-of-five 68%. A team that is a 65% favourite in each of its first three Swiss matches, the third of them a best of three, goes 3–0 only about 30% of the time.
Fixing results already played
Once a match is played, its result is locked in and only the rest is simulated. Swiss pairings that have already been drawn are used as drawn, so the odds sharpen as Worlds goes on. The ratings come from the season before the event, so results at Worlds enter the simulation as fixed outcomes, not as rating changes.
From odds to Pick’Em choices
- Play-In winner: the team with the best chance of winning it.
- Swiss 3–0: the two teams most likely to finish exactly 3–0.
- Swiss 0–3: the two teams most likely to finish exactly 0–3.
- Also advancing: the next six teams by chance of reaching the knockouts, skipping the four already picked.
If a pick’em asks for each team’s exact record, the odds table highlights the most likely one. On our own Worlds Pick’Em, each closed match shows the model’s pick and its chance of winning the series. The model is scored like any other player, so you can see how it is doing.
The Crystal Ball method
The Crystal Ball questions mirror the ones in the League client, and the answers rest on three building blocks.
Recent pro games, weighted. Champion questions use the last six patches of pro games, with each patch counting 15% less than the one after it. Champion win rates are steadied with five imaginary wins in ten games. Champions unlikely to play enough games at Worlds to meet a five-game minimum are left out. A patch-shift adjustment from ranked play covers the patches between the pro leagues and Worlds, as our tier list guide explains.
Expected games. Totals depend as much on how far a team goes as on how good its players are. To answer whether a player will reach a career kill milestone, for example, the model multiplies the player’s kills per game by the games his team plays in each simulated run, then works out the chance of getting there. Some player averages, such as KDA, are weighted by the chance their team reaches the knockouts.
The simulated brackets. Questions about how many regions reach the knockouts, or which of two teams places higher, are counted directly from the simulated events. Others run their own small simulations: the number of unique champions comes from redrawing whole pro series 2,000 times, because fearless drafting works within a series. Every answer comes with a breakdown of the numbers behind it.
What the model can’t know
- Roster changes. Team ratings follow results, so a team that changed players late in the season is rated partly on games played by a different lineup. The power rankings tab includes a roster rating that follows the players, but the event simulation uses team results.
- The patch. Worlds is often played on a patch the leagues never used, and a new meta can suit some teams more than others. Only the champion questions adjust for this.
- Form and preparation. Bootcamps, scrims, illness, travel and nerves all happen out of sight.
- Draft and sides. The model knows nothing about blue and red side, draft strategy, or a team saving picks for a best of five.
- Momentum. Each game is treated as independent, with the same chance as the last. Real series swing as teams adapt.
Read the odds as a well-informed starting point. A team with a 30% chance of reaching the semi-finals is a real contender, not a long shot. And if you know something the model doesn’t, trust it.