What people expect randomness to look like
Ask someone to write down a “random” sequence of coin flips, and they will usually produce something that alternates too often and avoids long runs. Real randomness is lumpier than that. It contains streaks, gaps and clusters, and they appear for the simple reason that each result is independent of the last.
When players see a run of misses, or three rare drops in a row, they often suspect the game is rigged or the odds are not what was stated. Sometimes they are right, but the existence of streaks alone is poor evidence. A sequence with no streaks at all would be the more suspicious one.
Counting streaks in a fair sequence
Consider a fair coin. The probability of five heads in a row starting at any particular flip is (1/2) to the fifth power, which is 1/32, or about 3.1 percent. That sounds uncommon.
But in a sequence of 100 flips, a run of five can start at any of the first 96 positions. At 96 windows with a 1/32 chance each, the expected number of five-head windows is 96 divided by 32, which is 3. Overlapping windows are counted separately, so a run of six heads contributes two of them, but the point stands: several runs of five or more are the normal case, not the exception.
A useful rule of thumb is that the longest run of identical results in n fair flips is typically around log base 2 of n. For 100 flips that is a little under seven, and the longest streak is often six or seven. A run of that length in 100 flips is unremarkable.
Streaks with unfair odds
The same logic applies to events that are not 50-50. Take a 20 percent crit chance. The probability of three crits in a row at a given starting attack is 0.2 cubed, or 0.008. Over 100 attacks there are 98 possible starting points, so the expected number of such windows is 98 × 0.008, just under 0.8.
At the other end, strings of misses are far more common. The chance of five consecutive non-crits starting at a given attack is 0.8 to the fifth power, about 0.33. A run of five attacks without a single crit is therefore a one-in-three event for any starting point, and across a long fight it should be visible several times.
This asymmetry is part of why streaks of bad luck feel more common than streaks of good luck for low-probability events: they are.
Gaps between rare events
Clustering shows up in a second way, in the spacing of rare events. Take an item that drops with probability 1 percent per attempt, independently each time.
- The chance the next drop comes within 10 attempts is 1 minus 0.99 to the 10th, about 9.6 percent.
- The chance you go 200 attempts without a drop is 0.99 to the 200th, about 13.4 percent.
So, for the same item, one player gets two drops within a few attempts and another waits over 200 without success. Neither player has been treated unfairly. The long-run average of one drop per 100 attempts is the same for both; the timing is not evenly spread, because random events do not space themselves out. The relationship between average and typical attempts is covered in the discussion of expected attempts.
Two drops close together and a long drought are two sides of the same effect. If the drops were spaced neatly 100 attempts apart, that would indicate a system that is not independent.
Why streaks feel meaningful
Several habits of perception make runs feel like signals.
Pattern detection. Humans are tuned to find patterns, including in noise. A cluster stands out and is remembered, while ordinary stretches are forgotten.
Selective memory. A player recalls the three-crit run in the last fight and not the forty normal attacks that surrounded it.
Post-hoc framing. After seeing a streak, we often ask how unlikely that specific streak was. But the right question is how likely it was that some streak of that kind would appear somewhere in the sequence, which is much higher.
Emotional weight. Missing a key shot several times in a row is frustrating, and frustration makes the event more memorable. This can feed into emotional tilt, where the perceived unfairness starts to affect decisions.
Two opposite mistakes
Seeing streaks produces opposing errors:
- The gambler’s fallacy: believing a streak must end soon because things should even out. A fair coin after six heads is still 50-50 on the next flip.
- The hot hand belief: believing a streak shows an underlying change and will continue. In a purely random system it does not. Whether a real hot hand exists in skilled performance is a separate question, and it is treated under the hot hand.
The two ideas point in opposite directions, yet both come from reading too much into a short run. A related confusion is overconfidence bias after a win streak, where a few wins lead to an inflated sense of ability.
When a streak is a genuine warning
Streaks are sometimes a sign of something else. Four things are worth checking before concluding it is just luck:
- Is the game using non-independent rolls? Some games alter odds after failures or successes. A pseudo random distribution produces fewer long droughts and fewer long hot streaks than independent rolls would.
- How many streaks are you looking at? If you watch many players, some will have very unusual runs by chance.
- Did you choose the window after seeing the result? Starting a count at the beginning of a streak inflates how remarkable it looks.
- Does the pattern continue on fresh data? A real effect should show up again; a coincidence usually fades.
A quick way to check a sequence
To see whether a run is plausible, estimate the probability of one specific run, multiply it by the number of places it could start, and compare the result to 1. If the expected count of such runs is around one or more, the run is ordinary. If it is tiny, such as a thousandth, then it is worth looking harder at the mechanics or the data.
Real random sequences are streaky. Learning to expect that, and to measure it, makes it easier to leave a normal bad stretch alone and to notice a genuinely unusual one.