The Kliwe rating
A number and a band: what we think you are worth, and how sure of it we are.
Where it sits
Elo gives you one number and moves it after every game. It is simple and it is immediate, but it cannot tell the difference between a rating built over four hundred games and one built over four.
Glicko fixes that by tracking how uncertain a rating is, but it does it in rating periods — you play, and some time later the number moves.
The Kliwe rating takes the useful half of each. It updates the moment a game ends, like Elo. It carries a confidence, like Glicko. And then it does two things of its own.
One: the band is public
Most sites compute an uncertainty and then hide it, showing you a clean number that pretends to a precision it does not have. We show it. Your rating is written 1200 ±350 on the day you arrive, and the band closes as you play.
That has a consequence worth stating plainly: the ranking sorts on the bottom of the band, not on the rating. A new player who wins five in a row has a high rating and a wide band, so they sit below a member with a lower rating that has been proven over a hundred games. You climb the table by removing doubt, not by getting lucky.
Below ±100 a rating counts as settled. Above it, it is shown with a ? and treated as the
guess it is.
Two: four ratings, not one
Bullet and classical are different games played with the same pieces, so each is rated on its own list — as FIDE keeps standard, rapid and blitz apart. Which list a game counts towards is decided by the clock, and FIDE decides that by how long a whole game is expected to take: the base time plus sixty times the increment.
| Category | A game is expected to last | Example |
|---|---|---|
| Bullet | under 3 minutes | 1+0, 2+0 |
| Blitz | 10 minutes or less | 3+2, 5+0, 10+0 |
| Rapid | over 10, under 60 minutes | 15+10, 30+0 |
| Classical | 60 minutes or more | 30+30, 90+30 |
That arithmetic has one consequence worth saying out loud: 10+0 is blitz here, not rapid. Ten minutes with no increment is exactly the blitz ceiling FIDE sets. Other sites draw the line elsewhere; we follow FIDE, because that is the line the rest of chess uses.
Your profile shows all four. The global ranking shows whichever you have proven best, and there is a separate table for each.
Three: there is nothing to farm
A rating gap counts for at most 400 points, whichever way it runs. Beating somebody far below you is worth close to nothing; losing to them costs you properly. Hunting weak opponents is simply a slow way of standing still.
What moves your rating
Three things decide how far one game moves you:
- Your own band. Wide means we are still working out where you belong, so you move in long strides — up to 52 points. Once it is narrow you move by a dozen at most.
- Your opponent's band. A result against somebody we are unsure of tells us less, so it counts for less.
- The gap. Beating a stronger player is worth more than beating a weaker one, as it should be.
Rust
A rating is a claim about how you play now. Time away from the board widens the band again — slowly, a few points a week — so a rating from two years ago is not treated as though it were from yesterday. Play a game and it tightens straight back up.
The rules in one place
| Starting rating | 1200 |
| Widest band | ±350 |
| Narrowest band | ±50 |
| Settled at | ±100 |
| Ranking sorts on | rating − band |
| Largest gap that counts | 400 points |
| Rating floor | 100 |
| Ratings kept | one per speed |