What it predicts
For every one of the 32 events at the Brussels Final, the model gives each contender a single number: their chance of finishing in the top three. It never names one winner.
That's deliberate. On the day, the fastest qualifier can false-start, get blocked in, or be caught on the line, so picking the exact winner is close to a coin toss between three or four names. Picking who makes the podium is a fairer question, and one you can check against the result afterwards. So every number on the site is about the top three, never the gold medal alone.
How the model learns
The model learns from the real podiums of every Diamond League Final, Olympic Games, World Championships, European Championships and Continental Cup from 2018 to 2025 (2020 was cancelled). These are actual results, taken straight from World Athletics, not anyone's opinion of who should have won. Each athlete is judged only on what they had done before that final was run, so a July final is never scored using an August race.
For each athlete in the running, the model works out 18 signals from their real season: their best mark this year and their career best, how steady they've been from meeting to meeting, which way their form is heading, how often they've raced, how long ago they set that best mark, how far behind the fastest in the field they sit, and their record head-to-head against this exact field. A random forest, a model that weighs many signals together, turns all of it into one number: the chance of a podium. It weighs them together because the signals affect each other. A superb season best counts for less, for example, if the athlete has barely raced all year.
And it's tested honestly, using walk-forward validation: the model is only ever trained on seasons before the year it's scored on, so the accuracy below comes entirely from Finals it had never seen. That's the difference between a real forecast and a model that has simply memorised the answers.
How accurate it is
Both numbers come from the same predictions; they just ask different questions. The first is the real job the site does: given the eight to ten athletes who actually make a Final, how often is the model's projected top three correct? The second is a deliberately harder test it never faces in practice. The two sit about twelve points apart, and neither is rounded up or taken from a flattering season.
Neither number knows who will turn up. Both are scored only on athletes who were on the start line, so they measure how well the model ranks a field, not who ends up in one. Injuries and late withdrawals are a separate question, and the flags on the projections are what answer it.
Where the data comes from
Every mark, ranking and result comes straight from World Athletics's own public API, the same data behind their broadcasts and athlete profiles. The scraping runs on a separate machine, and no mark is ever typed in or edited by hand, so what you read here is exactly what they published.
What it can't do
- It reads form, not the future. A last-minute injury, a withdrawal announced on the morning, or a slow tactical race decided by a final sprint can all beat the numbers on the day.
- It predicts who makes the podium, not the exact 1-2-3, and it never claims to know who wins.
- It's not affiliated with World Athletics or the Wanda Diamond League. It just reads their public data.
Searching by country
The search box finds countries as well as athletes. Type a nation — “Jamaica”, “KEN”, “Norway” — and it appears in the results alongside any athletes whose name matches, with how many ranked athletes and how many events that nation has this season.
Opening one gives that nation its own page: every ranked athlete it has, their marks and scores, its three best performances of the season, and whoever it has qualified for the Ultimate Championship. The page wears the country's own colours, taken from its flag.