How it works
You take over your favourite club for one match - you set the lineup and tactics, the computer replays your version of the match 5,000 times and checks whether you would have done better than the real team did that day.
The game loop in 3 steps
You take over the team
You pick a real match of your team and before kick-off you set: the eleven, the formation, the playing style and 2 planned substitutions.
The model plays your match 5,000 times
After the real final whistle the computer replays your version 5,000 times (Monte Carlo) and takes the most typical result. You are not guessing the real score - you are testing your version.
You get a result, commentary and points
Your version of the match, commentary in your chosen persona's style, a comparison with reality and points in the ranking. You compete with friends and other fans.
Where the points come from
You are not guessing the real result - this is not a prediction game. You earn points for smart, brave decisions that held up. Four questions, up to 120 pts in total.
- 50 pts
Was it better than reality?
Your version won or drew where the team actually lost (the goal margin counts too). This is the main component.
- 25 pts
Does it hold together?
The result has to be credible within your own simulation. An 8:0 fantasy sits below 1% in the distribution, so it scores nothing.
- 25 pts
Did you show courage?
Going against what the majority predicted - but only when something came of it. A pointless oddity is 0.
- 20 pts
Did the coaching work?
Your planned substitutions actually came on and improved the game. Good timing scores higher than a random change.
Full legend with step-by-step examples: Scoring.
Isn't this just randomness?
Isn't this just randomness?
“Random” does not mean “arbitrary”. A dice is random and dumb. This model is random and informed - it plays your match 5,000 times according to the strength of real teams and players.
- It is not a dice roll. Goals are drawn from a Poisson distribution, and each team's attack and defence are set by its strength (the Dixon-Coles model - the same one xG sites and bookmakers use). A stronger lineup really does win more often. The randomness is weighted by data, not flat.
- Why 5,000, not once. A single match is one sample - one roll tells you nothing about the odds. At 5,000 replays the median and the distribution stabilise (law of large numbers). Weather forecasts and election models work the same way.
- It is repeatable. The model is seeded - the same lineup gives the same distribution every time. It is not chaos, it is a fixed distribution. Run it again, you get the same thing.
- We show the distribution, we do not hide it. You see not just one result but how often it came up (e.g. “2:0 in 18% of replays”). The signal (median) and the uncertainty (spread) are out in the open. A bookmaker does the same, only sells you a single number.
- Luck does not score. A fluke result sits below 1% in the distribution - zero credibility points. And the ranking uses the average across many matches, so a one-off slice of luck averages out while skill stays.
- Player strength is not a hunch. The 0-99 rating is computed from season xG, assists and minutes. A change in the lineup genuinely shifts the goal distribution - and that is what we measure when scoring coaching decisions.
In short: we measure this randomness and show it. A single match can surprise you, but 5,000 replays plus a ranking on the average mean that over the long run the decision wins, not luck.
The full machinery under the hood (Dixon-Coles, Poisson, Elo, data) is described in the FAQ.
Ready to try?
Enter the World Cup, pick a national team or neutral mode and set a lineup for the first match. After the final whistle you will see how it would have gone with your version.