See it live

A prediction engine you can question

Our charts answer the questions we chose. Yours deserve their own.

We put real work into the charts and visualizations on this site. But every one of them answers a question we picked. Connect Predixsport to Claude and you can ask your own: custom analyses, custom charts, custom comparisons, drawn from the same models and the same published accuracy. We think this is the natural evolution of a sports analytics platform: not only a site you visit, but a prediction engine you can question.

Every chapter is a real question. Copy the prompt, paste it into Claude with the Predixsport connector, and get your own version with today's data.

One honest tip: ask Claude to build a chart and it will.

Chapter 1

This week's card, at a glance

Ask what is on and who the model favors. Claude reads the live win probabilities across both tennis tours and turns them into a chart. Below is the same data this site serves, the tightest toss-ups and the clearest favorites.

WTA Prague · hard
Lucie Havlickova Barbora Krejcikova
ATP Estoril · clay
Andrey Rublev Timofey Skatov
WTA Prague · hard
Nikola Bartunkova Lanlana Tararudee
WTA Hamburg · clay
Paula Badosa Panna Udvardy
WTA Prague · hard
Maria Timofeeva Sara Bejlek
ATP Kitzbuhel · clay
Mariano Navone Quentin Halys
WTA Hamburg · clay
Elina Avanesyan Yulia Putintseva
ATP Kitzbuhel · clay
Tomas Martin Etcheverry Ignacio Buse

A snapshot of the latest published card. On the live connector Claude pulls the current week.

Copy-paste prompt for Claude

What tennis matches are on this week and who does the model favor? Build me a chart of the win probabilities.

Chapter 2

Anatomy of a Grand Slam call

A prediction is only worth as much as the receipt. Here is exactly what the model published before the 2026 Wimbledon men's final, and exactly how it played out. The winner call was confident and correct. The length was not. We show both.

ATP Wimbledon 2026, men's final Pre-match forecast
Jannik Sinner Alexander Zverev
Most likely score 3-0 Sinner 37.4 percent of simulations
Expected total games 36.5 across the match
Full set-score distribution
3-0
37.4%
3-1
21.7%
3-2
12.3%
2-3
9.2%
1-3
11.3%
0-3
8.2%

Scores read from Sinner's side. The first three are Sinner wins, the last three are Zverev wins.

Why the model favored Sinner The factors that drove the call, in win-probability points. Positive favors Sinner.
Surface fit +8.87 Sinner
Overall record and opposition quality +6.14 Sinner
Recent record by level +5.64 Sinner
Overall strength +5.08 Sinner
Head-to-head +1.93 Sinner
Age +1.83 Sinner
Serve and return game +1.26 Sinner
Fatigue and recent workload +1.21 Zverev

Computed with gradient-based attribution on the neural network, not editorial opinion.

What actually happened
Winner Jannik Sinner Called right at 79.75 percent
Total games played 45 We expected 36.5. It went long.

An honest receipt. The winner was a confident, correct call. The length was a miss, the match ran eight and a half games longer than the model expected. We publish both, because a forecast you cannot check is just a claim.

Copy-paste prompt for Claude

What did you predict for the Wimbledon 2026 men's final, and how did it actually go? Build a chart comparing your pre-match forecast with the result.

Chapter 3

Who wins the whole thing

Match odds are one thing. Winning seven rounds is another. Our Monte Carlo engine runs every live draw thousands of times, so a favorite is not a name but a curve: the probability of surviving each round, decaying toward the title.

Survival curve, round by round · ATP · Estoril 2026
0%25%50%75%100%R16QFSFFTitleRublev 27.8%Blockx 22.1%Faria 10.9%Darderi 10.3%
Title probability, top contenders · WTA · Hamburg 2026
0%5%10%15%20%Badosa19.4%Bondar16.6%Oliynykova11.8%Kalinina11.3%Korpatsch10.7%Charaeva8.3%Putintseva6.2%Avanesyan4.3%

Rebuilt daily from 20,000 simulations of each open draw. The same numbers feed the live embed widgets.

Copy-paste prompt for Claude

Who is projected to win this week's tournaments? Chart the favorite's path round by round.

Chapter 4

Can you trust the numbers

A 70 percent call should be right about 70 percent of the time. That is calibration, and most sites never show it. Ask Claude to plot predicted probability against the observed win rate and you get this reliability diagram. Points on the dashed line are perfectly calibrated.

0.026 Expected calibration error
2,552 Held-out matches
64.9% Winner accuracy
Tennis winner model, held-out test set. Points close to the dashed diagonal are well calibrated.

The full set of reliability diagrams for every model lives on the model transparency page.

Copy-paste prompt for Claude

How accurate are your tennis predictions really? Plot predicted probability against observed win rate.

Chapter 5

The predictability duel

Some players are kind to a model. Others are chaos. We score each season by how often a player upset the model's own expectations. Ask Claude to compare two names and it builds the season-by-season story. Sinner against Alcaraz is the classic.

Season script, z-score per season (below zero = steadier than the model priced)
-3-2-10+1+2steadier than priced below the line-2.612024Straight sets-1.562025Straight sets-1.532026Straight setsJannik Sinner+1.922024Wild card-2.002025Straight sets-2.002026Straight setsCarlos Alcaraz
Season Jannik Sinner Carlos Alcaraz
2026 Straight sets z = -1.53 Straight sets z = -2.0
2025 Straight sets z = -1.56 Straight sets z = -2.0
2024 Straight sets z = -2.61 Wild card z = 1.92

Read down each column. Straight sets means the model's favorites mostly held. Wild card means more upsets than the model priced in. This is a per-season read, the gap does not carry across years.

Copy-paste prompt for Claude

Who is more predictable for your models, Sinner or Alcaraz? Build a season-by-season comparison.

Chapter 6

It works across sports

The same questions carry across everything we cover: NBA and Euroleague basketball, twelve football leagues, both tennis tours. Here is how our published predictions scored against two clubs over the 2025-26 season, measured the honest way, against the league average.

The 2025-26 NBA Elo race · Oklahoma City wire to wire, Boston from the cellar to contender (Elo re-seeds at 1500 every season, so one season per chart)
150015501600165017001750OctNovDecJanFebMarAprMayJunThunder 1734Celtics 1630
NBA Oklahoma City Thunder
On script

18.4 percent easier to call than the NBA average

Winner accuracy 78.9% league 67.4%
Brier score 0.1685 league 0.2064

2025-26 season, 90 predicted games. Lower Brier is better.

Football Inter
On script

42.9 percent easier to call than the Serie A average

Result accuracy 78.1% league 52.5%
Brier score 0.3538 league 0.6194

2025-26 season, 32 predicted matches. Lower Brier is better.

Copy-paste prompt for Claude

Chart the 2025-26 NBA Elo race between Oklahoma City and Boston, then tell me which team your model predicted best.

The prompt library

Use these prompts

Every prompt on this page in one place. Each one bakes in a chart instruction, so Claude builds the visual instead of just writing a paragraph. Copy, paste, and make it yours.

P1

What sports do you cover and what can I ask you?

P2

What tennis matches are on this week and who does the model favor? Build me a chart of the win probabilities.

P3

Pick the most interesting match on today's tennis card, explain why the model favors its pick, and turn the key factors into a chart.

P4

Who is projected to win this week's tournaments? Chart the favorite's path round by round.

P5

How accurate are your tennis predictions really? Plot predicted probability against observed win rate.

P6

Who is more predictable for your models, Sinner or Alcaraz? Build a season-by-season comparison.

P7

What did you predict for the Wimbledon 2026 men's final, and how did it actually go? Build a chart comparing your pre-match forecast with the result.

P8

Chart the 2025-26 NBA Elo race between Oklahoma City and Boston, then tell me which team your model predicted best.

Start asking your own questions

The Claude connector is free: 200 requests a month, no code, setup in a few minutes. Add it once and it works everywhere you use Claude.

Set it up on a computer, then it works on your phone too.