Flip a coin

How this coin flip works

Every flip asks the browser for a 32-bit random number through the crypto.getRandomValues() API. That number comes from your operating system's entropy pool — the same source your computer uses to create HTTPS session keys. Even values are heads, odd values are tails. There is no fixed seed, no pattern to discover, and no history that influences the next flip.

The distinction matters more than it looks. Math.random(), used by most coin-flip sites, is a pseudorandom generator: given the internal state, the entire future sequence is predictable. For deciding who buys coffee that is irrelevant. For a draw with an audience, money or a dispute attached, it is not.

Pick the coin you actually own

You can flip a US one-dollar coin or a Brazilian one-real coin. Both are drawn as vector graphics, so they stay sharp at any size and on any screen — the golden Sacagawea dollar with its eagle reverse, and the bimetallic real with its golden ring and silver core. The choice is cosmetic: the odds and the randomness source are identical. Your preference is saved in this browser.

When a coin flip is the right tool

• Two equivalent options. When both choices are equally good, the cost of continuing to decide exceeds the cost of being wrong. Flipping is the rational answer.
• Finding out what you wanted. A trick therapists know: flip the coin, then notice your reaction to the result. Relief or disappointment reveals the preference you were not admitting to.
• Setting order or side. Who goes first, who picks the field, who presents. The classic use, and it needs nothing beyond two possible outcomes.

When it is the wrong tool

Do not flip when the options have asymmetric consequences — one decision that could cost a lot and one that could not. Randomness is only fair between alternatives of similar weight. It is also wrong for decisions you can postpone and resolve with better information: the coin replaces the deadlock, not the research.

Physical coins are biased — this one is not

In 2023, a team of 48 researchers published the largest experiment ever run on the question: 350,757 hand flips using 46 different coins. The result was counterintuitive. Coins are not biased toward heads or tails, but they do tend to land on the same face that was up before the flip, about 51% of the time. The cause is physical: a human hand imparts a slight precession to the spin axis, so the coin spends more time with its starting face exposed.

In practice that means a live draw should hide the starting face or rotate the coin before tossing. Here the problem does not exist: there is no axis, no hand and no starting face. The probability is exactly 50%, flip after flip.

Privacy

The whole page works without a server. The draw, the tally and the history run in your browser and vanish when you close the tab. We do not record results, we do not require an account, and there is no usage limit.

Why your results look streaky — and why that is correct

Flip a coin twenty times and you will very likely see a run of four or five of the same face. People read that as the generator being broken. It is the opposite: a sequence without streaks would be the suspicious one.

In twenty fair flips, the chance of seeing a run of at least four in a row is roughly 77%. A run of five or more shows up about 47% of the time. Long runs are not rare events; they are what randomness looks like up close.

When researchers ask people to write down an imaginary sequence of coin flips, the fakes are easy to spot: humans alternate far too often and almost never write a run longer than three, because a real-looking run feels wrong. Genuine randomness has no memory and therefore no reason to balance itself out in the short term.

The gambler's fallacy, stated precisely

After six heads in a row, tails is not more likely on the seventh flip. It is still exactly 50%. The coin has no record of what it did before, and nothing carries forward from one flip to the next.

The confusion comes from mixing up two different questions. Before you start, the probability of seven heads in a row is 1 in 128 — genuinely unlikely. But once six heads have already happened, that part is settled and no longer a probability at all. The only open question is the seventh flip, and its odds are 1 in 2. The unlikely thing already occurred; it does not make the next event any less likely.

The counterpart is the "hot hand" version — believing heads is now more likely because it is "running". Both are the same error pointed in opposite directions: treating independent events as if they were connected.

What is true is regression toward the mean. Over 10,000 flips the ratio will sit very close to 50/50. But it gets there by the early imbalance becoming proportionally small against a growing total, not by the coin later producing extra tails to compensate.

Flipping for more than two options

A coin gives you one bit per flip, and you can spend those bits to choose between more than two things. Two flips give four equally likely outcomes (HH, HT, TH, TT), three flips give eight, and so on.

Choosing among three options is the interesting case, because three does not divide evenly into four. The correct method is rejection: flip twice, assign HH, HT and TH to your three options, and if you get TT, discard it and flip twice again. Every option keeps exactly a one-third chance, and the loop ends quickly — the chance of needing more than four rounds is under 0.5%.

The tempting shortcut of mapping TT onto one of the three options is exactly the wrong fix: it silently gives that option a 50% share instead of 33%. If you need to pick from a longer list, use the wheel instead, which handles any number of entries directly.