What your result means
The fan chart shows 2,000 copies of the same player, each playing the same number of games with different luck. The orange line is the median: half the copies are above it after each game, half below. The shaded band runs from the 10th to the 90th percentile, so eight in ten copies sit inside it. The thin blue lines are five individual copies, which is what one real player's history would look like. The dotted line is the average path, worked out exactly with no luck at all, for equal-rated opponents only.
The wobble is the standard deviation of the rating over the last 50 games, pooled across every copy. About two thirds of the time, a settled rating sits within one wobble of its middle.
What an Elo-style rating is
An Elo-style rating is a single number that stands for how strong a player is. Its key idea is that the difference between two ratings sets the expected result of a game between them. A 200-point advantage means an expected score of 0.76, so the stronger player should win about three games in four; a 400-point advantage means 0.91. Two players with the same rating are each expected to score 0.5.
After every game the rating moves by the K-factor times the gap between what happened and what was expected. Win a game you were expected to win and you gain a little; win one you were expected to lose and you gain a lot. The system never needs to know your true skill. It only keeps nudging the number in the direction your results point, and over many games that drags the rating towards the level where you win as often as the rating predicts. Many online games and chess-style ladders use variants of this rule.
Why new accounts climb quickly
A player whose true skill is 1800 but who starts at 1200 is matched against 1200-rated opponents. The rating thinks each game is a coin toss; in reality the player is 600 points stronger and wins 96.9% of the time. Each of those wins adds 16 points at K 32. In the simulation the copies win 93.0% of their first 20 games, so the rating shoots up at first.
The climb then slows. Every win lifts the player into stronger company, the real chance of winning drops towards one half, and the gain per game shrinks. That is why the curve bends: the last hundred points take far longer than the first. With the defaults the average path is within 50 points after 63 games but needs 98 games to get within 10.
Why K matters: a fast climb or a steady rating
The K-factor is the most points a single game can move the rating. A big K reaches the true level quickly: from 600 points below, the average path gets within 50 points in 31 games at K 64, 63 games at K 32 and 127 games at K 16. The second chart shows the whole curve from K 8, which needs 255 games.
The price is noise. Once the rating has arrived, every game is close to a coin toss, and each result still moves it by about half of K. For a player whose rating starts at their true skill of 1500, the wobble is 37.3 points at K 16, 53.4 at K 32 and 75.8 at K 64. Doubling K multiplies the wobble by about 1.4. A rating system has to choose between reacting fast and keeping still, which is why some ladders use a large K for new players and a smaller one later.
Why matchmakers want equal-rated opponents
A game between two players the system rates as equal carries the most information: either result is a genuine surprise to the rating, so it moves by the full K/2. A game against someone rated 300 points lower barely moves it when you win, and costs a lot when you lose. With a matchmaking spread of 300 points the median takes 88 games to get within 50 points instead of 61, and the copies win only 86.7% of their first 20 games instead of 93.0%. Once settled, the wobble hardly changes: 53.0 points against 53.2.
How it works: the formula
Ropp = R + m, m ~ Normal(0, spread)
win with probability p(S − Ropp)
Rnew = R + K × (result − p(R − Ropp))
average path: Rn+1 = Rn + K × (p(S − Rn) − 0.5)
- p(d) the expected score, here the win probability, for a rating difference of d points.
- S the player's true skill, constant for every game. R the displayed rating, starting at R0.
- Ropp the opponent's displayed rating. The matchmaking spread is the standard deviation of the gap; at 0 every opponent has exactly your rating. Each opponent's true skill is assumed equal to their displayed rating.
- result 1 for a win, 0 for a loss. Draws are not modelled.
- K the K-factor. The average path is the rating with luck taken out, exact for a spread of 0 only.
- Monte Carlo: 2,000 runs from a seeded random-number generator (mulberry32), so a link with the same seed always gives the same chart. The random gap uses the Box–Muller method.
Worked example: the default ladder
True skill 1800, starting rating 1200, K 32, every opponent rated the same as you, 150 games.
- Game 1: the rating expects 0.5; the real chance of winning is p(600) = 0.969. A win adds 32 × (1 − 0.5) = 16 points.
- The average path adds 32 × (0.969 − 0.5) = 15.02 points in game 1, then less every game: 1709.23 at game 50 and 1791.30 at game 100.
- It gets within 50 points of 1800 after 63 games and within 10 points after 98.
- Across 2,000 seeded runs (seed 1), the median rating is 1,712 at game 50, 1,776 at game 100 and 1,808 at game 150, and the median first gets within 50 points at game 61.
- Over the last 50 games the rating wobbles by ±53.2 points.
How to use this simulator
- Set how good the player really is and where the ladder puts them at the start. Move the sliders, or tab to one and use the arrow keys.
- Change K to trade a faster climb against a noisier rating, and widen the matchmaking spread to see what uneven opponents do.
- Run again moves to the next seed: a fresh set of 2,000 players with new luck. Copy link saves every setting, the seed included. Save picture downloads the fan chart with the settings and headline printed underneath.
Starting gap, K-factor and matchmaking in numbers
The tables come from the page's exact average-path recursion, its exact first-game calculation and its seeded Monte Carlo (2,000 runs, seed 1), so they match what the page shows with the same settings.
| Starting rating | Gap to true skill | Real chance of winning game 1 | Games to within 50 points | Games to within 10 points |
|---|---|---|---|---|
| 1,700 | 100 points | 64.0% | 15 | 50 |
| 1,600 | 200 points | 76.0% | 31 | 65 |
| 1,500 | 300 points | 84.9% | 41 | 75 |
| 1,400 | 400 points | 90.9% | 49 | 84 |
| 1,200 | 600 points | 96.9% | 63 | 98 |
| 1,000 | 800 points | 99.0% | 76 | 110 |
The page's default, 600 points below true skill, wins its first game 96.9% of the time and needs 63 games to get within 50 points and 98 to get within 10. Starting only 100 points below needs just 15 games to get within 50, since there is far less distance to cover.
K-factor
| K-factor | Games to within 50 points | Games to within 10 points | Wobble once settled |
|---|---|---|---|
| K 8 | 255 | 394 | ±25.7 |
| K 16 | 127 | 196 | ±37.3 |
| K 24 | 85 | 130 | ±45.9 |
| K 32 | 63 | 98 | ±53.4 |
| K 48 | 42 | 65 | ±65.4 |
| K 64 | 31 | 48 | ±75.8 |
Doubling K roughly halves the games needed (255, 127, 63 and 31 for K 8, 16, 32 and 64) and multiplies the wobble by about 1.4. A K of 16 settles to ±37.3 points, a K of 64 to ±75.8.
The path
| K-factor | Game 10 | Game 25 | Game 50 | Game 100 | Game 150 |
|---|---|---|---|---|---|
| K 16 | 1,274 | 1,379 | 1,530 | 1,708 | 1,771 |
| K 32 | 1,345 | 1,531 | 1,709 | 1,791 | 1,799 |
| K 64 | 1,476 | 1,712 | 1,792 | 1,800 | 1,800 |
The average path is not a straight line. At K 32 it climbs 145 points in the first 10 games and only 7.9 points between games 100 and 150. At K 64 it is already within 10 points by game 50 (1,792).
Uneven opponents
| Spread of opponents | From a start of 1200 | From a start of 1600 |
|---|---|---|
| None | 96.9% | 76.0% |
| 100 points | 96.4% | 74.5% |
| 200 points | 94.8% | 71.4% |
| 300 points | 91.9% | 68.2% |
A wider matchmaking spread lowers the real chance of winning early games, because stronger opponents appear sometimes: from 1200 the first-game chance falls from 96.9% at no spread to 91.9% at 300 points. The page notes this slows the climb and leaves the settled wobble almost unchanged.
Six mistakes when reading a rating ladder
- Choosing the biggest K. A bigger K reaches the true level sooner but wobbles more once there. The table has K 64 settling in 31 games and wobbling ±75.8 points.
- Reading the rating as exact skill. At K 32 a settled rating wobbles by about ±53 points from luck alone. Differences smaller than that do not mean a change in skill.
- Expecting a fixed number of games. The number depends on the gap and on K, from 15 games for a 100-point gap to 76 for 800 points, at K 32.
- Assuming real ladders use this rule. Many add draws, a step that shrinks as a rating becomes certain, placement games, bonuses and floors. The page models only the plain update.
- Treating the average path as what happens to one player. The average path is exact for equal opponents only. A single player's line is noisy, and the fan chart on the page shows the spread.
- Forgetting that skill is held fixed. The model assumes the player never learns, tires or tilts, and that opponents' ratings are exactly right.
Assumptions & limitations: what the model leaves out
- No draws: every game is a win or a loss.
- One fixed K. Many real systems also track how uncertain a rating is and shrink the step as it settles; that is not modelled.
- Opponents are rated exactly right, and their ratings do not move. In a real ladder everyone is also climbing or falling.
- The player's true skill never changes: no learning, no off days, no tilt.
- Solo games only, with no team ratings, placement matches, bonuses or floors from any real service.
Frequently asked questions
How many games does it take to reach my true rating?
It depends on how far away you start and on the K-factor. In this model, a player 600 points under their true skill at K 32 needs 63 games for the average path to get within 50 points and 98 to get within 10. At K 16 it takes 127 and 196 games.
Why does a bigger K-factor make the rating wobble more?
Once a rating has arrived, each game is close to a coin toss and still moves it by about half of K. In this model a settled rating wobbles by about 37 points at K 16, 53 at K 32 and 76 at K 64.
Why does a new account climb so fast at first?
Because it is playing opponents far weaker than its true skill. With the defaults the player is 600 points stronger than the first opponent, wins 96.9% of the time, and gains 16 points per win at K 32. The gains shrink as the opponents get stronger.
Does uneven matchmaking slow the climb?
Yes. With a matchmaking spread of 300 points, the median rating takes 88 games to get within 50 points instead of 61 with equal-rated opponents. The settled wobble barely changes, 53.0 points against 53.2.
Is this how a real game ranks its players?
Not exactly. Many online games and chess-style ladders use variants of this update rule, but most add things this model leaves out, such as draws, a step size that shrinks as a rating becomes more certain, and placement games. Treat the numbers as an illustration of how the rule behaves.
Related calculators
Sources and references
- Fenner, Levene and Loizou, Birkbeck, University of London: A Discrete Evolutionary Model for Chess Players' Ratings (arXiv 1103.1530) — the expected-score curve with the 400-point scale, the update rule new rating = old rating + K × (score − expected score), K as the most a single game can move a rating, and a larger K giving more weight to recent results
