The Desk's year-end chances came true about as often as they said
The Desk's year-end chances came true about as often as they said. Of 4,520 players given eight to nine in ten, the outcome happened 86.3% of the time, so a high chance meant what it said. The forecast gives each player a chance that their NTRP rating moves up, stays the same, or moves down, and every chance on a player page comes from it.
A chance is never a promise for one player.
How close were the chances?
Close for every group of players but one. Each row groups players by the chance the forecast gave its most likely outcome. Of the 28,976 players given nine in ten or more, the forecast said 97.0% on average, and the outcome happened for 97.3%.
| Chance the forecast gave | Said, on average | Happened |
|---|---|---|
| Five to six in ten | 55.0% | 52.9% |
| Six to seven in ten | 65.3% | 65.4% |
| Seven to eight in ten | 75.3% | 74.1% |
| Eight to nine in ten | 85.6% | 86.3% |
| Nine in ten or more | 97.0% | 97.3% |
| Mixed only, a move | 74.0% | 40.9% |
What about players in Mixed doubles alone?
Their forecasts were the one exception. For the 137 of them whose most likely outcome was a move, the forecast ran high, because it said a move far more often than one came. So their player pages now print the share that came true for players like them.
All Tennis Desk Research·Browse every league the Desk holds, by year and area
How this was measured
Scope. The Desk scored 39,855 players in the USTA NorCal leagues it covers, at the 2023, 2024 and 2025 year-ends. Each year-end's forecast learned only from earlier year-ends, so it never saw the results that the Desk checked it against.
The check. The Desk compared each forecast with the NTRP rating the player held the next season. Players with no rating the next season do not count, because there is no new rating to compare.
Measured. The Desk checked every figure above against the live database on November 17.
Method. The Desk produced every figure above with the SQL in docs/research/records/forecast-findings.json, and ran it twice to the same result before it built this page.