A chess rating is a single number that predicts something quite specific: the likely result between two players. Understanding what it does and doesn't say is worth more than chasing it.
Rating systems are built around one relationship — the difference between two ratings predicts the expected score, not the winner.
Roughly, under Elo:
| Rating gap | Stronger player's expected score |
|---|---|
| 0 | 50% |
| 100 | ~64% |
| 200 | ~76% |
| 400 | ~92% |
Two things fall out of that table, and both matter when a stake is involved.
A gap of 200 points is not a guarantee. It's roughly three wins in four. Losing one game in four to a weaker opponent is the system working as designed, not evidence that you were cheated.
Small gaps are nearly coin-flips. At 50 points apart, the stronger player scores about 57%. If you're picking opponents, the edge you feel is usually smaller than the edge you have.
Elo (after Arpad Elo, a person — it isn't an acronym) is the original: you gain points for wins, lose them for losses, and the amount depends on how surprising the result was. Beating someone 300 points above you moves your rating a lot; beating someone 300 below barely moves it.
Its weakness is that it treats all ratings as equally trustworthy. A brand-new player's 1500 and a veteran's 1500 after two thousand games are numerically identical but epistemically very different.
Glicko (and Glicko-2) fixes that by tracking a second number alongside the rating: rating deviation, a measure of uncertainty. A new or returning player has high deviation — the system knows it doesn't know — so their rating moves in large steps until it settles. An active player with a long history has low deviation and moves slowly.
Most modern platforms use Glicko-style systems for exactly this reason, including ChessBit.
Because ratings are only comparable within one pool. A rating is a position relative to the other players in that system, not an absolute measure of strength. Different sites have different player populations, different starting ratings and different rating floors, so a 1600 in one place can be a 1400 or an 1800 in another. Comparing across platforms tells you almost nothing; comparing your own trend within one platform tells you a lot.
Ratings are also per time control. Your bullet and rapid ratings measure genuinely different skills and routinely differ by hundreds of points — which is why platforms keep them separate rather than averaging them into a fiction.
A provisional rating is one the system doesn't trust yet: too few games, too much uncertainty. During that period your rating swings hard, because each result carries a lot of information.
On a free site that's a curiosity. On a real-money platform it's a fairness problem in both directions:
- An unrated newcomer might be a titled player, and their first opponents shouldn't be the ones who find out at their own expense.
- Equally, a genuinely new player shouldn't be matched against someone far stronger while the system works out where they belong.
This is why ChessBit asks new players to play calibration games before staking, and why matchmaking pairs within a compatible rating band rather than at random. The mechanics are in matchmaking and ratings.
- Watch the trend, not the number. Fifty games is a signal; five is noise.
- Don't chase it. Playing weaker opponents to protect a rating is the surest way to stop improving.
- Use it to pick your stakes, not your ego. The right stake is the one you'd be relaxed about losing four times in a row — because at a 200-point gap in your favour, that will still happen sometimes.
ChessBit is a skill-based real-money chess platform in early access — 18+. Responsible play.
