Mallorca Open, March 2026. I am running between the courts.
"Who made the final against Kyle and Paq?"
"The Belgian brothers."
For a second I think: Levi and Yosha? They are not here.
"Who?"
"The Paes brothers."
That is how I learned they exist.
Six months later, in Paris, Martijn and Daan Paes 🇧🇪 beat Nelson and Rose 🇺🇸, the new world champions. The model gave them 9.8%.
It was not even the biggest upset of the week.
First group round of the men's individual. Jewitt and Killas 🇬🇧 against Cote and Gagnon 🇨🇦. One game to 17, no second chance. The British pair won.
By Federico Grazi's rating model, that win had a 5.4% chance. No result in Paris was won at a lower probability.
So I read the results the other way round. Not who won. Who was not supposed to.
Federico, thank you. You built the model and you keep the archive. Without both, this would be my gut feeling 🙂
My yardstick is Federico's model, the one that priced the open division before Paris. For every game it can price, it gives each side a probability of winning.
Two limits. Only 1,146 of the 2,763 Paris games in the archive carry a probability, and mixed is not in the file at all.
One term for newcomers. In a squad event two nations meet in a tie of five pairings, each one called a rubber, and the nation with more rubbers wins the tie.
1. How unlikely was the biggest one?
What does 5.4% feel like? Put a chip on a split at a European roulette table, two numbers out of 37. That is 5.4%.
You would not bet your rent on it. It still lands.
The model rated Cote and Gagnon 498.5 points higher than Jewitt and Killas. In Guildford 2024 the least likely finished win came at 12.2%.
Jewitt and Killas won that group 4-0. Cote and Gagnon went through in second place at 2-2.
In the second group round Jewitt and Killas landed in a Golds group, the tier the eventual champions played in, and lost all five games there.
Same pair, same week.
I know James Killas. I played him once, in advanced mixed at the Munich Masters in June 2024. Two levels below Contender and Pro. I remember him super fast, running everywhere.
This year he climbed. His first Pro divisions in the archive, Dresden and Kiel, 7 wins from 9 at the Mallorca Open in March, and a career-high 1,491 before Paris.
2. Can you win the upset and still lose the tie?
Ask Belgium 🇧🇪. Women's squad, round of 16 against Canada 🇨🇦, best of three games.
Dockx and De Coster beat Yu and Dorval 2-0 at 6.0%. Lauwers and Delbecke beat Morehouse and Cunningham 2-0 at 10.5%.
Canada won the other three rubbers and the tie 3-2. Belgium went out.
3. Who else got under 10%?
Schra and Schippers 🇳🇱 beat Yu and Dorval 🇨🇦 at 9.0%, in a women's squad consolation game for places.
Sejersen and Weber 🇩🇰 beat Regard and Pecorino 🇨🇭 17-12 at 9.3% in the second group round of the women's squad. Switzerland won the tie 4-1 anyway.
4. And the Paes brothers?
At 9.8%, Martijn and Daan Paes 🇧🇪 beat Nelson and Rose 🇺🇸 in the second group round of the men's individual.
It was the only match Nelson and Rose lost as a pair in the men's individual on their way to the title.
5. Why does Canada keep showing up?
Ten of the 25 least likely wins came against Canadian pairs 🇨🇦. Cote and Gagnon, and Déry and Prince at 10.3%, in the men's individual. Forand and Vallee twice, in women's individual placement games. Six in the women's squad.
That is a count, not a rate. I have not matched every Canadian game, so this is a reading of the list, nothing more. Why it happened, I cannot tell you from this file.
And I know these players personally. This list rates results. It does not describe the people.
6. Which one does not count?
The raw list has Golaszewski and Rawiak 🇵🇱 over Mercadini and Guerra 🇮🇹 at 10.8%. It is not an upset.
Italy had already won the first three rubbers of that men's squad round of 16, so the last two were stopped. The archive books this one, stopped at one game to none, as a Polish win.
So is the 11.1% entry atop the 2024 list. My ranking leaves out every stopped rubber. The overall counts below keep them as recorded.
7. So was Paris full of upsets?
"But Jakub, five wins under 10%, that sounds like a wild week?" Not really.
In Paris the model priced 1,137 decided results. Set aside the two it rated dead even, and the favourite won 889 of the remaining 1,135, 78.3%. The model had those favourites at 74.6% on average.
In the 228 games where one side was below 10%, that side won five, against about 13 expected.
Is the model a little too careful with favourites, or did Paris simply go to form? Hard for me to say from one week.
Which result from Paris surprised you most, and is it on this list? Tell me at contact@futureroundnet.com.
FRESH is the Future Roundnet camp in Mallorca, for players from advanced level up. If you want to hear when the next edition opens, leave your name under Keep me posted at futureroundnet.com/fresh-27.
The ratings and probabilities come from a model built and run by Federico Grazi. I am quoting the model, not vouching for it. Selection, nation matching and errors are mine.
Corrections: spot something wrong? Write to contact@futureroundnet.com and I will fix it and say so here.
Quick guide: the world championship in brief
By Federico Grazi's rating model, Jewitt and Killas of Great Britain beating Cote and Gagnon of Canada in the first group round of the men's individual. The model gave them a 5.4% chance.
If two pairs rated exactly like these played that game a hundred times, the model would expect the underdog to win about five. It describes how rare a result is, not whether it can happen.
More long-shot wins, and more long-shot games: in Paris five of 228 games with a side below 10% went to that side, in Guildford none of 107. In both years the favourites won more often than the model expected, 78.3% against 74.6% in Paris and 79.9% against 71.5% in Guildford.