The problem with short runs
Imagine flipping a fair coin ten times and seeing seven heads. Does that suggest the coin is biased? Many people would say yes. The arithmetic says otherwise.
With a fair coin there are 2^10 = 1,024 equally likely sequences of ten flips. The number of sequences with exactly seven heads is 120, with eight heads 45, with nine heads 10, and with ten heads 1. Together, seven or more heads can happen in 120 + 45 + 10 + 1 = 176 of the 1,024 sequences, which is about 17 percent. That is roughly one run of ten flips in six. Nothing is wrong with the coin; this is simply what small samples look like.
Games work the same way. A sample size of ten rounds, ten hands or ten matches is too small to separate skill, method or a lucky streak from ordinary variation.
How the spread shrinks as trials grow
What gets more stable with more trials is the proportion, not the raw count. Take the coin again.
- In 100 flips, the number of heads typically varies by about 5 either side of 50, so a result between 45 and 55 is unremarkable. That is a range of 45 to 55 percent.
- In 1,000 flips, the typical variation is about 16 either side of 500, so a result between roughly 484 and 516 is unremarkable. That is about 48 to 52 percent.
The typical size of the swing follows a square-root rule: the spread in the count is half the square root of the number of flips for a fair coin. For 100 flips that is 0.5 x 10 = 5, and for 1,000 flips it is 0.5 x about 31.6, which is roughly 16. To halve the proportional uncertainty you need about four times as many trials. This is why more data helps, but with diminishing returns.
The general principle is the law of large numbers, which tells us that long-run averages approach the true probability. It does not say that streaks get cancelled out or that you are due a correction.
What this means for judging progress
Suppose you are trying a new approach to a game and you win more than before. Is it working? Ask a few questions.
How many games is that? If the answer is a dozen or two, the honest conclusion is that you cannot tell yet. A good run and a real improvement look the same in a small sample.
What did you expect before you started? If you decide in advance how many games to play and what you will compare, you avoid stopping at a flattering moment. Looking only at the best stretch of a long record is a form of cherry-picked statistics, even when you do it to yourself.
How large is the difference? A big, consistent gap is more convincing than a slim one. A slim gap needs far more games to separate from noise. Confidence intervals are the formal tool for putting a margin around a win rate, and even a rough sense of them stops you from overreading small changes.
Is the opposing side comparable? Beating weaker opponents for a while and then facing stronger ones changes what a win rate means.
Streaks look meaningful but often are not
In long sequences of random events, runs appear more often than most people expect. A run of five wins or five losses is not unusual once you play enough rounds. After an extreme result, the next ones tend to look more ordinary, a tendency known as regression to mean. That is not a force pulling results back; it is what happens when an unusually good or bad run was partly luck.
For that reason, beware of changing your method after a few bad results or sticking to a lucky one. Both reactions treat noise as signal.
Fair ways to track progress
- Pick a minimum number of games before drawing any conclusion. Choose it ahead of time and stick to it.
- Compare like with like. Track results by format, opponent level or stake, so a change in conditions does not masquerade as a change in skill.
- Look at the process as well as the result. Counting how often you made a sound choice is more stable than counting wins.
- Use rolling averages rather than single results. The last 50 games say more than the last 5.
- Expect to wait. If the effect you are looking for is small, the honest answer for a long time may be “undetermined”.
What this does not change
A long record can tell you about your own play, but it cannot turn a negative game into a positive one. In games with a built-in house edge, more trials move your results closer to the expected loss, not away from it. More play is therefore not a way to fix a bad average, and there is no point at which a loss run “has to” reverse. If you play with money, decide the amount beforehand, never borrow to continue, and reach out to a local support service if you feel you are losing control.
The takeaway for experience
Understanding sample size makes games calmer to play. A bad evening is rarely evidence that you have declined, and a good one is rarely proof of mastery. Judging results by their quantity as well as their size lets you enjoy the game without being pushed around by every swing.