How the rating works

You are unrated until four matches are recorded. A best-of-three counts as one, a single set as a half, so eight single sets get you there too.

Until then nothing is published. No provisional number, no placeholder, no rating with an asterisk. It is the same way a golfer has no handicap index until enough cards are in.

The scale runs 1.0 to 7.0, the same one padel uses worldwide. It is a fixed formula rather than a position in a queue, so your number does not move because somebody else joined.

One full level apart means the stronger player wins about three times in four. Half a level is closer to three times in five. That is what the number is promising, and it is the thing to check it against.

Below is the same 6-4 6-3 scoreline, played by a settled 3.5 player. The only thing changing is who was on the other side of the net.

Look at the top row. Beating a 1.5 pair only 6-4 6-3 is a worse result than your rating expected, so it earns you nothing. Winning never costs you points. It can be worth zero, which is what stops anybody climbing by picking on weaker opponents, but a match you won will never drag you down.

Losing can still earn you something, and the bottom rows show it: take a set off a pair two levels above you and your rating goes up, because you did better than anyone expected.

Your expected result is worked out from the two pairs, not from you against their pair. Put a weaker player alongside a much stronger one and the pair is favoured, so a win is roughly what was expected and moves the number very little.

The reverse holds too. Lose alongside a weaker partner to a pair you were expected to beat, and it costs you properly, because the pair should have won.

None of which is a reason to avoid better players. Play with them, learn from them, enjoy it. It simply means the rating reflects your own play rather than the company you keep. We measured the difference: how strong your partners are predicts your rating at 0.76 without this correction and 0.05 with it, in a simulation of 200 players over a season. It is the single biggest reason this is a rating rather than a league table.

Half of each result is simply who won. The other half is how comfortably, measured in games. A 6-0 6-0 says far more than a 7-6 7-6, and ignoring that would mean needing many more matches before your rating meant anything.

One blowout cannot throw you across two levels, though. The margin is capped.

Your rating is worked out from your most recent 30 matches, the way a golf handicap uses your recent rounds rather than your whole life. Play better than you used to and the old form stops arguing.

It is a count, not a calendar. Stopping for a season costs you nothing: come back and your rating is exactly where you left it, because those matches are still your most recent ones. What does change is how sure we are: after a long gap your first few results move you further than usual.

Qualifying is separate. The four matches that earn you a rating count every match you have ever played, so nobody loses a published rating by being away.

A tournament result is entered by the organiser who watched it, so it counts in full. Everything else is entered by one player and agreed by an opponent, which is good evidence but not the same thing, so it counts for 70% as much.

Format matters the same way and for the same reason. A best-of-three is a bigger test than a single set, so a single set counts half, both toward your rating and toward the four matches that publish it. One rule, not two.

Ratings are never nudged up and down and left there. The whole match record is replayed from the beginning each time, so if a score is corrected, a match is voided, or two duplicate profiles are merged, every affected rating repairs itself on the next run.

It also means we can always show you exactly which results produced your number.

Two questions: roughly what level you are, and who you already play. Neither answer is ever shown to anyone, including the people you name.

They do different jobs. Your own estimate gives us somewhere to start. Naming players we may already have results for places you relative to them, which is far more precise. On their own, each is mediocre; together they are close to accurate before you have hit a single ball.

None of it survives contact with real results. Four matches in, the maths has largely replaced whatever you told us.

Everything above is tuned against simulated players, because Barbados has not produced enough recorded matches yet. The ordering, meaning who is better than whom, which is what seeding actually needs, is robust. The exact mapping onto the 1.0 to 7.0 scale is a calibration we will re-fit once a few hundred real results are in.

When we do, every historical rating is recalculated with it, and we will say so publicly rather than quietly changing the numbers.

The full working, covering what is proven, what is assumed and what would be hard to defend if you challenged it, is written up in the repository as docs/rating-audit.md.