Too Even to Be Random
A teacher catches the students who faked their coin flips with one rule. Real randomness is lumpier than you think.
Your music app shuffles to a song, and then — the very next track — another one by the same artist. “This shuffle is broken,” you mutter. You’re not alone: enough people complained about clusters like this that Spotify and Apple both quietly redesigned their shuffle to be less random, precisely so it would feel more random. The true shuffle kept doing something our brains refuse to accept: it clumped.
This is one of the deepest and most useful facts about randomness, and almost nobody’s intuition gets it right. Real randomness is lumpy. It streaks, it clusters, it leaves gaps. The neat, evenly-spread pattern we picture when we hear the word “random” is, ironically, the one thing randomness almost never produces. And the gap between what randomness actually does and what we think it should do is so reliable that you can use it to catch a liar.
A teacher gives every student the same homework: flip a coin 100 times and write down the results, H or T, in order. She suspects some students will be lazy and simply invent a string of H’s and T’s out of their head instead of actually flipping.
When the papers come back, she catches nearly every faker in seconds, using a single simple rule — without watching anyone flip. What is she looking for?
Give it a real think before reading on. Most people’s first instinct is exactly backwards. We assume the faked sequences will be the ones that look “too patterned” — and that the real coin flips will be a nice, healthy, well-mixed jumble of heads and tails. So we go hunting for something that looks suspicious.
But when a person sits down to fake randomness, they do something very human: they spread things out. They avoid repeating themselves. After writing HHH, it feels wrong to write a fourth H — surely a real coin would “correct” itself? So they switch. They alternate more than they should. Their fakes come out looking suspiciously tidy, suspiciously balanced — too even to be random.
The teacher’s trick keys on exactly the thing people can’t bring themselves to fake. Figure out what that is, then open the interactive: play a round of “spot the real one,” try to fool a randomness detector by tapping out your own sequence, and finally scatter points across a field and watch true randomness build clumps and empty voids all on its own.
Explore this puzzle visually with an interactive diagram — drag sliders, watch the geometry update in real time, and build intuition before you solve.
Don't ask "which sequence looks more patterned or suspicious?" — that instinct points the wrong way. The real coin flips are the ones that will look, to your eye, a little too wild.
Think about what a person does when they invent a fake sequence from their head. After writing three of the same letter in a row, what does the fourth one feel like it should be? People almost can't bring themselves to keep a streak going, because a long run "looks unrandom" to them.
So don't count how balanced the heads and tails are. Instead, find the longest unbroken run of the same face in each sequence — and ask which is longer. Real randomness produces long streaks constantly; fakers suppress exactly the thing that gives them away.
She looks for the longest streak. In a real run of 100 coin flips, there is almost always a stretch of six, seven, or even eight of the same face in a row. The fakers, who couldn’t stomach writing more than three or four in a row, give themselves away by the absence of a long streak. The tidy paper is the fake one.
The numbers are lopsided enough to make this nearly foolproof. In 100 honest flips:
- A run of 5 or more in a row appears about 97% of the time — it’s almost guaranteed.
- A run of 6 or more shows up around 81% of the time.
- The single longest streak, averaged over many sequences, is about 7.
Meanwhile, ask people to write a fake sequence and their longest streak is typically just 3 or 4. That’s the tell. The honest sequences are full of streaks that look rigged; the rigged sequences are suspiciously free of them.
Why the streaks are there
Each flip is independent — the coin has no memory. After five heads in a row, the chance of a sixth is still a plain one-half; the coin doesn’t owe you a tail. And across 100 flips there are so many places for a streak to start that a long one becoming almost inevitable. (A rough rule of thumb: the longest run in n flips tends to be about log₂(n) — and log₂(100) is about 6.6, right in line with what we see.) What feels to us like a bizarre, patterned coincidence is just chance doing its ordinary, lumpy work.
The same lumpiness lives in space, not just in time. Scatter stars, raindrops, or bomb craters at random across a map and they won’t spread out politely — they’ll form apparent clusters and bare patches. During the 1944 flying-bomb attacks on London, residents were certain the bombs were targeting particular neighborhoods because the hits appeared to cluster. The actuary R. D. Clarke divided South London into a grid and checked: the pattern fit pure randomness almost perfectly. The “clusters” were the fingerprints of chance, not of aim.
This one idea explains a surprising amount
Once you see that real randomness clumps, a whole family of everyday errors clicks into place:
- The gambler’s fallacy: believing a run of reds at roulette makes black “due.” It doesn’t — the wheel, like the coin, has no memory.
- The “hot hand”: seeing a basketball player’s streak of made shots as proof they’re heating up, when streaks of that length are roughly what pure chance would produce anyway. (This one is genuinely debated — more below.)
- The clustering illusion: reading meaning into the coincidental clumps in cancer maps, lucky numbers, or a shuffled playlist.
The honest caveat
None of this means clusters never matter. Sometimes a cluster really does have a cause — a contaminated well, a genuine hot streak, a real bias in a coin. The lesson isn’t “ignore all clusters.” It’s subtler and more powerful: a cluster by itself is not evidence of a cause, because randomness manufactures clusters for free. Before you claim a pattern is meaningful, you have to ask whether plain chance could have produced it — which is exactly what Clarke did with the bombs, and exactly what the “hot hand” researchers did with the shooters. That question is what separates a real signal from a trick of the eye. (It’s worth knowing that a 2018 reanalysis found a subtle statistical bias in the original hot-hand studies, nudging the debate back toward the players — a lovely reminder that measuring randomness well is harder than it looks.)
There’s a beautiful physical echo of all this. In 1827 the botanist Robert Brown watched pollen grains in water jitter about with no cause he could name; in 1905 Einstein explained the dance as the sum of countless random molecular kicks. The path such a particle traces — a “random walk” — doesn’t drift smoothly outward; it wanders, doubles back, and clumps, straying from its start only about as fast as the square root of the time. Whether in a coin, a playlist, a scatter of bombs, or a mote of pollen, the signature is the same: real randomness is lumpier than we ever expect.
One email, one puzzle, no noise — with a hint ladder and a full worked solution.