No simple averages.The story the scoreboard leaves out.
A raw score is one panel's opinion on one day. We correct every published score for how its meet judged, read her whole season, and place her against every gymnast in the country on one ruler. This page shows what that changes, with the numbers.
Every published USAG score, every level, Development Program and Xcel. Bias-corrected, not bias-free. Patent pending.
Her season, two ways
Her raw average says 8.75. Her season says 8.86, because she climbed all winter and no meet judged her the same way twice.
Emma is a sample gymnast · the judging and the numbers are illustrative
We do the hard math. So what?
Four answers, each measured on every published score in the country. Averages are not wrong. They are incomplete, and this is what they leave out.
How much does the judging move a score?
The same routine scores about a quarter point higher at a generous panel than at a tight one, so we measure every panel and take the wind out of every score.
Why isn't her average her number?
A season climbs about a point from the first meet to the last, so an average lands in the middle of a story that is still moving; her Flip Score reads where she is now, from her own history.
Does it actually change the rankings?
Rank the country by raw average and by Flip Score, and the middle gymnast moves by hundreds of places at the big levels; a quarter of the raw top 100 is not in the corrected top 100.
Why does a tenth matter so much?
In the crowded middle of a big level, a single tenth is hundreds of places, which is why a judging swing of a quarter point cannot be left in the number.
One ruler. Six answers.
Because every score sits on the same ruler, we can place her against the country, write her a goal from her own history, and show what happened to every gymnast who stood where she stands. These are the six things you get that no average can give.
Where does she sit on the curve?
Her Flip Score on the national curve for her exact level and event. An average has no curve to sit on.
Is she on pace for her goal?
A goal written from her own history before the season, one per path, locked at her first meet. Not the level standard.
Is she near the level-up lines?
Two lines from national history for her level, and where she sits between them. Her coaches decide.
Who walked her path?
Every gymnast who stood where she stands, and what each did next. Counts of real seasons.
Where do our lineups rank?
Three season Flip Scores per level, added up, ranked in the state and percentiled nationally. Not the meets your gym happened to attend.
Do gymnasts improve more here?
Every gymnast against her own goal, pooled across the roster, against every published gym. Growth, not talent.
Every picture is a sample gymnast or a sample gym. In the product each one carries an ⓘ that says what it means, what it doesn't, and what you can say.
Four ideas, no math degree.
The short version. The longer one is the next section down.
Every routine in a session flies in the same wind. A generous panel is a tailwind, a strict one a headwind. We measure the wind from the gymnasts who competed there and had history elsewhere, and take it out of every score. Never more than a quarter of a point, typically far less.
Her Flip Score reads her whole season and weights where she is now, from her own history. One down meet does not become her number, and a real jump shows up within a couple of meets.
A 9.0 on Level 6 floor and a 9.0 on Level 9 beam are different achievements. The Flip Rating puts every level and event on one 0 to 1000 scale, where 500 is the national middle of her own level. Top 3% means the same thing everywhere.
Two lines from national history for her exact level: where half of gyms would move up a gymnast like her, and where gymnasts usually finish in the top half at the next level. The chart says where kids like her tend to be. It never says what to do.
The method, in more detail.
You clicked through, so here is more of it than the sales pages carry: where the math comes from, what goes in, how we grade ourselves, and the words you will meet. The constants and the weights stay ours. The method is patent pending.
Where this math comes from.
None of it was invented for gymnastics. It is the same family of math that other fields use when the readings are noisy, the thing being measured keeps moving, and the samples are small.
Navigation never trusts one reading. It carries a belief about where you are, takes the next noisy fix, and moves part of the way toward it: further when the fix is clean, less when it already knew. Her Flip Score works the same way on each event. It carries a belief about what she can score, watches the next meet, and moves part of the way, less for a bouncy event like beam than for vault, less for a veteran with fifteen meets on record than for a newcomer with two. One counted fall dents an established number. It does not redefine her.
A forecast blends what the model believed yesterday with what the instruments say today, and it is never a single number: it comes with how sure it is. So does every number here. Her Flip Score is published with its own uncertainty, which narrows as meets come in, and the goals and the comps are ranges with a middle, not a point.
A sprint record is read with the wind taken out. Every score here is read with its judging panel taken out, and the panel is measured from the gymnasts who competed under it and already had history elsewhere. Only gymnasts with a record vote. A fall does not vote, because we do not blame judges for gravity. And a panel's number is pulled toward zero in proportion to how few votes stand behind it, then capped: never more than a quarter of a point, typically far less.
An insurer does not trust a tiny pool, and a growth chart places one child against the whole country at her age. Both habits are here. Every gym's number is pulled toward the national middle in proportion to how little evidence it has, so a hot five-gymnast gym cannot leapfrog a proven forty-gymnast program on luck. And the level-up lines are drawn from every gymnast at her exact level, the way the lines on a growth chart are drawn from every child.
No simple averages. No straight lines.
An average treats her first meet in November and her last in March as the same day, and treats every panel as the same panel. A straight line through her scores assumes she improves at a constant rate and will keep doing so. Neither is true of a season. Her number is allowed to bend with time, weights where she is now, and never fits a line through her scores. The goals are built the same way: not as a score, which drags every gymnast toward the average, but as a change from where she is, with gymnasts who stay at a level and gymnasts who move up read as two different populations, because they are. Where a move up usually costs a gymnast something, that cost is measured from the thousands who made the same move at her score tier, not guessed.
What goes in: dozens of measurements, in four families.
Every one of them is about her, her room, or the country's history. The readiness model alone weighs more than forty of them. What we do not publish is how much each one counts; that is the recipe.
- Every published score, and each meet's judging panel
- The level and the event, and the days between meets
- How settled her number is, and which way it is moving
- Her best, her consistency, her momentum over the last few meets
- The move she made between levels, and what that move usually costs
- Her teammates at her level, measured with her own scores left out
- The strength of her gym at that level
- How tough her state's field is
- How crowded her meets were
- Where she sits on the national distribution for her level and event: the percentile, the median, the spread around it
- What gymnasts who looked like her did the following season
- What a move up cost gymnasts at her score tier, by quarter of the field
- Where in the season she is, and how many meets are in
- The first-meet effect: a first meet is usually a gymnast's worst
- Which seasons count most: recent ones, by design
How we grade ourselves.
A number is only as good as the test it survived. These are the tests, run on seasons the models never saw.
Words you will meet here.
Corrected, not perfect.
One fair standard. Every meet.
The scores were always free. What they mean is the product.