Flip a coin once, or ten thousand times. Each flip is a bit from the browser's cryptographic generator, so it is genuinely fair — but the more useful part is what the tool shows you about runs. People are remarkably bad at recognising randomness: in a hundred fair flips a run of six the same way is more likely than not, and nearly everyone who sees one concludes the coin is weighted. The longest run of each side is shown explicitly rather than hidden, because a tool that displays a history without mentioning this quietly reinforces exactly that mistake.
| In 100 flips | Reality |
|---|---|
| A run of 5+ | Very likely |
| A run of 6+ | More likely than not |
| Exactly 50/50 | Only about 8% of the time |
| 58–42 split | Completely ordinary |
| After 5 heads | Next flip is still 50/50 |
Yes. Each flip is one bit from crypto.getRandomValues, the browser's cryptographic generator, with no seed you could predict and no state carried between flips. A real coin is arguably slightly less fair than this, since the side facing up when it is flipped has a small edge.
No, and this is the most common thing people ask. In a hundred fair flips a run of at least six is more likely than not. Runs feel wrong because our intuition expects randomness to alternate far more than it does — which is why the longest streaks are shown here rather than hidden.
No. The coin has no memory, and nothing about the previous flips changes the next one — it is 50/50 again, exactly as it was the first time. That belief is the gambler's fallacy, and it is the reason a history display without this caveat does more harm than good.
Because it should not be. In a hundred flips, 58–42 is entirely ordinary. The proportion converges toward half as you flip more, but the absolute gap tends to grow — flip ten thousand and you will likely be within a percent proportionally while being dozens of flips apart in absolute terms. That is what the law of large numbers actually says.
Up to ten thousand. Doing so is the clearest way to see the proportion converge while the absolute difference widens.
No. The history lives in the page until you leave it.
Each flip is a single bit from the browser's cryptographic generator, drawn by rejection sampling for consistency with the other tools here even though a two-way split divides evenly and cannot be biased by a modulo. The part worth building is not the flip, it is the presentation of runs. People are extremely bad at recognising what randomness looks like: asked to write a fake sequence of coin flips, almost everyone alternates far more than chance does, and shown a real sequence containing six heads in a row, almost everyone concludes the coin is weighted. In a hundred fair flips a run of at least six is more likely than not. So the longest run of each side is shown explicitly rather than buried — hiding it would let the tool quietly reinforce the belief that a streak means something. The running tally is shown for the same reason and with the same caveat: a hundred flips landing 58–42 looks lopsided and is entirely ordinary, since the expected deviation grows with the square root of the count while the proportion converges only slowly. The page says outright that previous flips carry no information about the next one — the gambler's fallacy is the single most common misunderstanding about this, and a tool that displays a history without saying so is inviting it. Bulk flipping is offered up to ten thousand at once, which is enough to watch the proportion converge while the absolute gap widens, and that is the clearest demonstration of what the law of large numbers actually claims. Nothing is transmitted.