Methodology
For every player ranked 80–400, the model outputs a calibrated probability of reaching the ATP Top 100 within 12 and 24 months. Calibrated means the number behaves like a betting odd: of players we mark 30%, roughly three in ten make it. No composite "AI scores" without a testable meaning.
Every list is produced under strict time-slicing: the model trains only on data available before the list's publication date, then the following 24 months score it. We roll this forward across every season 2015–2024. Nothing in the training window can see the future it is judged on. Full year-by-year results on the track record page.
Match results and weekly rankings across tour, Challenger and ITF levels since 1990; audience data collected from public social platforms. Sources are licensed for commercial use in production.
Injuries are inferred from schedule patterns, not medical data. Sample sizes below age 19 are small and flagged. Probabilities are estimates with error bars, not guarantees — we publish the 80% interval on every forecast.
Methodology — Targets are calibrated probabilities of reaching the ATP Top 100 within 12/24 months. Training data: tour, Challenger and ITF results plus weekly rankings since 1990. Backtests are strictly time-sliced; training always ends before the prediction window opens.
Preview build — players, figures, probabilities and prices shown are illustrative sample data.
© 2026 AgenTennis