Main event · September 18
Playoff predictions
Eight teams, fourteen matches, double elimination. The model reads the bracket from what teams showed at this very International — and from the season before it. You set the balance: the “TI form” slider moves the estimate from pre-tournament to tournament.
- Sample 185 maps · 56 TI series
- Simulations 60,000
- Teams in bracket 8
- Draft closes August 20 · 05:00 AM
Playoffs
Your bracket
Click a team to send it forward. The winner moves along the bracket, the loser drops to the lower one; matches further down the branch that became impossible are cleared. The percentages in a card are the odds of winning that exact pairing. “Fill from the model” picks the bracket with the most expected points, not the strongest teams: sometimes it pays to back whoever reaches the match more often.
Upper bracket
quarterfinals
-
M1 August 20 Bo3 -
M2 August 20 Bo3 -
M3 August 20 Bo3 -
M4 August 20 Bo3
semifinals
-
M5 August 21 Bo3 -
M6 August 21 Bo3
Upper bracket final
-
M7 August 22 Bo3
Grand final
-
M14 August 23 Bo5
Lower bracket
Round 1
-
M8 August 21 Bo3 -
M9 August 21 Bo3
Round 2
-
M10 August 22 Bo3 -
M11 August 22 Bo3
Round 3
-
M12 August 22 Bo3
Lower bracket final
-
M13 August 23 Bo3
This is the model bracket — edit it and it becomes yours
Team odds
Team odds
Probabilities from the bracket simulation. They move with the model settings and after every finished series.
Team Spirit 7–3 · 16–10 +138 100% - 100
- 0.0
- 0.0
- 0.0
- 0.0
- 0.0
0.0 Maps ahead
Iron Wing 4–3 · 11–8 +10 0.0% - 0.0
- 0.0
- 0.0
- 0.0
- 0.0
- 100
0.0 Maps ahead
TEAM VISION 7–1 · 16–8 +78 0.0% - 0.0
- 100
- 0.0
- 0.0
- 0.0
- 0.0
0.0 Maps ahead
BoomBoys 5–4 · 12–11 +9 0.0% - 0.0
- 0.0
- 0.0
- 100
- 0.0
- 0.0
0.0 Maps ahead
Team Liquid 5–3 · 11–10 +3 0.0% - 0.0
- 0.0
- 0.0
- 0.0
- 100
- 0.0
0.0 Maps ahead
Team Yandex 5–5 · 14–13 +3 0.0% - 0.0
- 0.0
- 100
- 0.0
- 0.0
- 0.0
0.0 Maps ahead
Nigma Galaxy 5–3 · 12–7 +61 0.0% - 0.0
- 0.0
- 0.0
- 0.0
- 100
- 0.0
0.0 Maps ahead
Team Falcons 4–4 · 12–11 −33 0.0% - 0.0
- 0.0
- 0.0
- 0.0
- 0.0
- 100
0.0 Maps ahead
How many maps the team still plays on average across simulations
What the tournament changed
What the tournament changed
The season estimate knows everything about a team except how it plays right here. A hundred maps is too little for a standalone rating, but enough to move the seasonal one.
- Team Spirit 44 maps at the event
- In meta
- 47%
- Heroes
- 34.8
- Ban exposure
- 44%
- Season
- 1732
- TI shift
- +138
- Tempo
- 0.20
- Meta
- 0.19
- Result
- 1873
- TEAM VISION 38 maps at the event
- In meta
- 48%
- Heroes
- 29.8
- Ban exposure
- 46%
- Season
- 1794
- TI shift
- +78
- Tempo
- -0.19
- Meta
- 0.41
- Result
- 1875
- Nigma Galaxy 28 maps at the event
- In meta
- 45%
- Heroes
- 23.8
- Ban exposure
- 43%
- Season
- 1663
- TI shift
- +61
- Tempo
- -0.40
- Meta
- 0.28
- Result
- 1725
- Iron Wing 26 maps at the event
- In meta
- 50%
- Heroes
- 23.1
- Ban exposure
- 46%
- Season
- 1707
- TI shift
- +10
- Tempo
- 0.19
- Meta
- 0.27
- Result
- 1721
- BoomBoys 34 maps at the event
- In meta
- 55%
- Heroes
- 31.0
- Ban exposure
- 48%
- Season
- 1743
- TI shift
- +9
- Tempo
- -0.09
- Meta
- 0.85
- Result
- 1760
- Team Liquid 28 maps at the event
- In meta
- 52%
- Heroes
- 24.4
- Ban exposure
- 51%
- Season
- 1723
- TI shift
- +3
- Tempo
- 0.27
- Meta
- 0.67
- Result
- 1734
- Team Yandex 37 maps at the event
- In meta
- 58%
- Heroes
- 32.0
- Ban exposure
- 49%
- Season
- 1750
- TI shift
- +3
- Tempo
- 0.08
- Meta
- 1.84
- Result
- 1772
- Team Falcons 29 maps at the event
- In meta
- 29%
- Heroes
- 25.8
- Ban exposure
- 37%
- Season
- 1748
- TI shift
- −33
- Tempo
- 0.15
- Meta
- -1.72
- Result
- 1699
Event meta
Event meta
A breakdown of every played draft: who gets picked, who gets banned and who wins. A ban counts as much as a pick — a hero nobody is allowed to take defines the event no less than the one everybody takes.
How the prediction works
How the prediction works
- 1
Season strength
A rating over two years of maps discounted by match age and corrected for schedule strength, recent form, results on the current patch and market odds.
- 2
TI form
A second estimate is fitted on the tournament maps: the same Bradley–Terry model with the season strength as a prior. The trust slider moves the width of that prior — from “ignore the event” to “almost forget the season”.
- 3
Tournament corrections
Tempo — how fast a team closes the maps it wins and how long it holds the ones it loses. Meta — how well its hero pool matches what works on this patch, here.
- 4
Bracket simulation
Twenty thousand runs over the real Valve tree: every match is drawn from the map probability, the winner and the loser travel along their own edges. Finished series are not re-drawn, ongoing ones continue from the current score.
- 5
Prediction points
Your bracket is checked in every run: as many matches guessed, as many points by the Valve table. The scale is convex, so the safe bracket with the most hits is not always the best one.
The tournament sample is 185 maps across sixteen teams, about seven each. That is too little for a standalone rating: a single series moves the estimate by tens of points. This is why the season is never fully dropped, even at the top of the slider, and the uncertainty band is widened.
This is a model, not a prophecy. It shows how often an outcome happens under the stated assumptions and knows nothing about injuries, roster moves or what a team cooked up at a bootcamp.
