Coins and dice have no memory, yet results drift toward half. What a 1% chance really means, the hot hand illusion and the birthday problem.
Five heads in a row does not change the probability of heads on the next toss; it stays one half. A coin has no mechanism for storing past results. People nonetheless feel that balance must be restored and judge from that feeling. This is the gambler's fallacy. Provided the coin is fair, a streak tells you nothing about the next result.
It is true that the ratio approaches half over many tosses. That happens not because past imbalance is compensated but because, as trials accumulate, the early imbalance becomes a smaller share of the whole. The run of five does not disappear; it is diluted. Separating those two explanations changes how you judge.
At 1%, it feels as though a hundred attempts should produce one hit. In reality, the chance of getting nothing in a hundred attempts is over a third. Low-probability trials carry both a high expected count and far wider variance, so planning from the average alone goes badly wrong.
Shown a genuinely random sequence, people feel it is too clustered to be random, because real randomness naturally produces runs. Sequences people invent to look random contain too few runs. A result feeling strange is therefore not evidence that anything was manipulated. Indeed, when people are asked to invent a sequence of coin tosses, they almost never write long runs of the same side.
The usable principles are simple: do not predict the next result from the last, plan low-probability attempts around a near-worst case rather than the average, and set a limit before starting. This explains how probability works; it is not advice about which games to play or how much.
There is also an illusion pointing the other way from the gambler's fallacy: believing that a run of successes means a lucky streak that will continue. Some research suggests such streaks may partly exist where skill and condition matter, as in human sports, but in independent trials such as coins, dice and gacha pulls there is no such thing. Predicting the next result from past ones is the same error in either direction.
How many people need to be in a room before the chance that two share a birthday exceeds one half? Many people guess over a hundred, but in fact 23 is enough. Intuition quietly swaps in the question of finding someone with your birthday, whereas the real question accepts any two people matching, so the number of pairs to compare grows far faster than the number of people. Many apparent coincidences are, in the same way, more common than they seem.
Probability sinks in faster by doing than by formula. Toss a coin twenty times and record it, and you will see runs of four or five of the same side turning up more often than you expected. Repeating the same draw many times with a random picker and counting the results also works well. Watching a ratio that jumps around over a few trials settle near the theoretical value as the count grows leaves the difference between average and spread as a lasting sense.