It climbs a tenth of a point at a time. Across 4,493 computer-rated player-seasons of Marin-area and San Francisco USTA NorCal adult league, a player sitting at the top of their rating held a higher one the next year 21.5% of the time.
A tenth of a point under the top it was 7.1%, and four tenths over it 94.1%. There is no line to cross, only a slope to sit on.
A player counted as moving held a different published NTRP rating the next year, and that is the whole of it. The Desk cannot separate a promotion earned over a season from a self-rating being corrected, an appeal, or a reclassification, so none of these numbers is a forecast about any particular player.
An NTRP rating covers a range of playing strength rather than a single number. A 4.0 covers everything above 3.50 up to and including 4.00, so the top of that rating is 4.00 and every tenth above it is a tenth over.
The Desk reads a level off match results to two decimals, which is what makes a tenth of a point a step worth measuring. Everything below is measured from that top edge.
This page used to say that finishing above the range was followed by a move up 39.3% of the time. True, and close to useless.
That figure averages a stretch running from 21.5% at the top edge to 94.1% four tenths past it. Two players on either side of it are in different worlds, and one number for both of them tells neither of them anything.
The climb is steep and it is orderly, and a player can place themselves on it to the nearest tenth, which is a far more useful thing to know than where a boundary sits.
At a tenth of a point a single rating runs out of players. Two to four steps of eight carry enough for any one rating, and self-rated players over their rating carry enough nowhere.
So the curve is measured across every rating together, and across computer-rated players only. That is a real limit, and I think it is the price worth paying for the resolution.
The coarse view is still true underneath. 20.1% of every move up came from a player who never passed their rating, and 54.1% of every move down from one who never finished below it.
More than 0.30 under the top of the rating, 11.0% of computer-rated players moved down, and by a tenth closer in it is already 1.2%.
Self-rated players are the ones who fall, and they fall from places where a computer-rated player holds steady. A self-rated player who finished low inside the range moved down 22.4% of the time, against 4.8% for a computer-rated player in the same place.
Some of that gap is not tennis at all. A self-rating can be corrected without anybody playing badly, and a correction looks exactly like a move down in data that only sees the published rating.
Four tenths of a point separate 21.5% from 94.1%.
Free, no signup. The Tennis Desk reads the real USTA NorCal match record.
All Tennis Desk Research·Browse every league the Desk holds, by year and area
Grounded — every figure above is a count over a stated denominator. No estimates, no model output: just match results and what they add up to.
Scope. USTA NorCal adult league for Marin County and San Francisco, seasons 2022–2025, each followed into the next year's published rating. It is those two areas, not NorCal and not a national figure.
Who is counted. A player-season enters only if the same person carries a published NTRP rating in both years — a move is not observable otherwise — and only if the Desk holds ten or more of that player's lines. That second rule is about the Desk's coverage, not about how much anyone played: the crawl covers two areas, and a player seen twice here may have played a full season somewhere it does not reach. It is not a USTA minimum, and nothing here shows that one exists. These line counts also include Combo, which USTA excludes from year-end ratings entirely. 5,224 player-seasons clear both.
Where a player finished. Read from the Desk level after their last recorded line of the season, against the range their published rating names. Seasons whose rating carried an appeal or a disqualification are excluded: they behave like neither group, and every one of their cells is too small to publish. I hope to measure them properly if coverage reaches the rest of NorCal or another section.
Method. Every figure above was produced by the SQL recorded in docs/research/records/where-do-you-have-to-finish-to-get-bumped.json and re-run against the live database before this page was built. A group under fifty player-seasons, or with fewer than five moves, publishes its size instead of a rate. Reproduce with npx tsx scripts/measure-research-record.ts.