How the rating system works, why online and FIDE ratings do not compare, and what level of play each band genuinely represents.
Elo is probably the most watched and least understood number in chess. Understanding it properly has an immediate practical effect: it stops hurting so much, and becomes a tool rather than a verdict.
Elo does not measure how much you know about chess. It measures one thing only: the probability that you will beat another player with a known rating.
That is the entire definition. It is not a grade, not a certificate and not a measure of talent: it is a statistical prediction, corrected after every game.
The idea is simple and needs no mathematics:
Step 1. Before the game, the system computes an expected result from the rating difference. If both players are equal, 0.5 is expected — draws, or half the wins. If one is 200 points higher, they are expected to score about 76%.
Step 2. Afterwards, it compares expected with actual. Win = 1, draw = 0.5, loss = 0.
Step 3. The difference between those two numbers is multiplied by a K factor, and that is what gets added or subtracted.
From that come the consequences everybody knows in practice:
It is how much your rating moves per game. In FIDE:
The logic: while the system does not yet know your strength, it moves fast to find your level. Once it has many games, it moves slowly so as not to react to noise.
Online sites use similar systems — usually Glicko — which additionally account for how confident the system is about your rating: if you have not played for months, your first result moves you far more.
This is the most common confusion and it is worth being clear: they are different scales and there is no official conversion.
A rating is always relative to the population that shares it. Two separate systems can assign very different numbers to the same player simply because they started from different initial values and never mixed.
In practice, an online rating usually sits well above the same player's FIDE rating, and by how much varies by site and by time control. It is not worth translating with a formula: it is worth looking at each scale separately.
And there is an additional difference that matters: FIDE ratings are earned in long games against federated players; online ratings, mostly, in fast games against a far wider and more varied population.
As an approximate reference, on an online scale:
These bands are indicative and vary by site. They are for orienting yourself, not for arguing.
Elo is not the only system, and platforms use different things under similar names. It is worth knowing which one you are looking at, because the numbers do not compare:
Glicko and Glicko-2. Used by most of the big sites. The difference from Elo is that alongside your score they keep a deviation: how much confidence the system has in that number. A new player has a high deviation and their score moves a lot; somebody who has played for years has a low one and moves little. That is why your first twenty games move a hundred points and the next twenty move ten.
FIDE Elo. The official one, calculated by tournament and updated in batches. It is the slowest and the most comparable across countries.
National ratings. Every federation has its own, calibrated on its own population. An 1800 from a small federation and an 1800 from a large one are not the same thing, and neither equals an 1800 FIDE.
The practical rule: a rating is only comparable with itself. Your number three months ago on the same platform at the same time control is the only comparison that means anything.
It is the commonest confusion and it has three causes, all technical:
The starting point differs. Each site decides what a new player begins with, and that decision shifts the whole scale. If one site starts everybody at 1500 and another at 800, the numbers will never match even with an identical population.
The population differs. A rating measures your strength relative to whoever plays there. A site with many strong players produces lower numbers for the same real strength.
The time control differs. Bullet, blitz and rapid have separate ratings for a good reason: they are partly different skills. A two-hundred-point gap between your blitz and your rapid is normal and means nothing bad.
From which comes the answer to the question everybody asks: there is no reliable conversion table. Approximations exist, they change over time and by site, and none is stable enough to base a decision on.
Almost everybody goes through a three- or four-month stretch where the number does not move. Before concluding you stopped improving, three checks:
Are you playing the same time control? Switching from rapid to blitz lowers the number without anything having got worse.
Who are you playing against? A stable rating while facing steadily stronger opponents is hidden improvement. Pairings rise only when you rise, so staying level costs more every month.
How many games have you played? With ten games a month, statistical noise is larger than any signal. A bad month with few games says absolutely nothing.
And if all three answers are in order and the number is still flat, then there is something concrete to look at, and it is almost always the same thing: you are working on the part you already know. A player doing a thousand puzzles a month and analysing no games is improving something that is no longer holding them back.
The fastest way out of that stretch is to stop asking the rating and start asking your games: twenty losses, grouped by cause, say in half an hour what the rating does not say in six months.
Do not look at it every day. It moves thirty points on statistical noise. Checking daily is reading noise and feeling it as signal.
Look at the three-month trend. That is the minimum window where the number says something.
Do not change plan because of a streak. Five losses in a row is a normal event, not a diagnosis. The diagnosis comes from analysing the games, not from looking at the graph.
And remember what it measures. Your rating does not say how much you know or how much you improved: it says how likely you are to beat someone with a similar number. You can have improved a great deal in three months and have the same rating, because what improved has not yet turned into results.
That happens constantly, and it is the number one reason people quit right before the jump.
How many points do I gain beating somebody much stronger? Almost the maximum your K factor allows, because the system expected you to lose. With K=20, beating somebody four hundred points above gives about 19 points; losing to somebody four hundred below costs the same.
Why does my rating fall even though I win more than half? Because the system does not count games won but expected score. Winning sixty per cent against opponents averaging a hundred points below you lowers your rating: more was expected.
How many games before the number is reliable? About thirty against opponents of similar level. Before that the number moves a lot by design, because the system does not yet know where to place you.
Is my online rating any use for knowing my real strength? It is useful as an internal reference and not as an equivalence. The only thing it says with certainty is how you do against that site's population, at that time control, right now.
If you want to put this into practice with structured material:
Four classes and eight hours of training to master modern evaluation, read the engine, handle complex positions and improve your endgame technique.
ARBy GM Andrés Rodríguez
A six-stage course to master pawn structures, develop effective attacks and make better decisions in demanding positions.
ARDFBy GM Andrés Rodríguez and GM Diego Flores
A collection of six courses by GM Andrés Rodríguez to enrich your repertoire, understand complex positions, and improve your game from opening to endgame.
ARBy GM Andrés Rodríguez
The six real differences between playing on a screen and playing across a table, and why so many people play considerably worse in their first tournament.
Read the articleWhat to bring, how a round works, what to do with the clock and the scoresheet, and the differences from online chess nobody warns you about.
Read the articleWhy they show up, why you should not try to make them disappear, and five concrete things that reduce calculation errors under pressure.
Read the articleARDFBy GM Andrés Rodríguez and GM Diego Flores
