Is an Online Coin Flip Actually Fair?

What “fair” actually means

A fair coin flip has two properties: each outcome has exactly a 50% probability, and each flip is independent of the ones before it — the coin has no memory. Nothing about “fair” guarantees a particular short run looks like an even split. That’s the part most people get wrong, and it’s the source of nearly every “is this coin flipper rigged?” question.

How software flips a coin

A digital coin flip is a random number generator picking between two equally likely outcomes — functionally, drawing a single random bit. Done properly, there’s no physical asymmetry to worry about: no weight distribution, no edge geometry, no air resistance. The Coin Flipper draws from a proper random source so the 50/50 claim holds by construction, not just approximately. Which random source matters less than you’d think here: a cryptographically secure generator and a seeded pseudorandom generator can both produce a genuine 50/50 distribution over many flips — the difference between them is about predictability, not fairness, and a seeded flip stays fair while remaining reproducible if you know the seed (more on that below).

Why 10 flips looking 70/30 is completely normal

This is the single most common source of “the coin flip feels broken” complaints, and it’s pure statistics.

Flip a fair coin 10 times. The expected split is 5-5, but “expected” describes an average over many sessions, not a guarantee for any one. The standard deviation of the proportion of heads in 10 fair flips is about 0.158 (15.8 percentage points), so a 70/30 split is only about 1.3 standard deviations from 50/50 — well within normal variation. Concretely: the probability of seeing a split at least as extreme as 7-3 in 10 flips is roughly 34% — about one in three ten-flip sessions will produce a 70/30-or-worse split purely from chance. Humans are bad at intuiting this; we expect small samples to already look like the long-run average, and they usually don’t.

The law of large numbers, with numbers

The fix for small-sample noise isn’t a “better” coin — it’s more flips. As flip count grows, the proportion of heads converges toward 50%, even though the absolute gap between heads and tails can still grow in raw count. The standard deviation of the observed proportion shrinks with sample size: about 15.8 points at 10 flips (70/30 unremarkable), about 5 points at 100 flips (70/30 now a 4-standard-deviation event, essentially never happening by chance), and about 1.6 points at 1,000 flips (even 55/45 starts looking unusual).

“Flip more” always beats “eyeball ten flips and draw a conclusion.” Running a few thousand trials and watching the running proportion settle is more convincing than any explanation — the Probability Experiment tool is built for that, and the Distribution Sampler generalizes the idea beyond binary outcomes.

Physical coins are slightly unfair too

Here’s a genuinely surprising fact: a real, physical coin toss isn’t perfectly 50/50 either, and the effect is well-studied, not folklore. Research led by Persi Diaconis (building on earlier physics-of-tossing work with Susan Holmes and Richard Montgomery) predicted that a hand-flipped coin has a small “dynamical bias” — it’s slightly more likely to land on the same face it started on, due to precession during the toss rather than weight imbalance. A large-scale empirical follow-up, hundreds of thousands of tosses across many people and coins, found a same-side bias near 51%, close to the original prediction.

Two caveats: the effect is small, about 1 percentage point above chance, undetectable without many careful, tracked tosses; and it depends on the starting position being tracked and not re-randomized, which casual flips rarely do. It’s real, but not a reason to distrust an ordinary flip day to day — if anything it argues for software, where a well-built digital flip has no analogous mechanical bias to correct for.

How seeds make flips verifiable after the fact

One thing a physical coin can’t easily give you: proof, after the fact, that the outcome wasn’t tampered with, without a trusted witness watching. A seeded digital flip can, via a commit-reveal protocol: publish a hash of the seed before the flip, flip using that seed, then reveal the seed afterward. Anyone can recompute the flip and confirm it matches the announced outcome — and since the seed was committed before the result was known, there’s no way to have quietly swapped in a more favorable one. This is the same idea behind “provably fair” schemes on some online gaming platforms.

When not to trust any coin flip

For genuinely high-stakes decisions — real money, legal or contractual outcomes, anything where one side has an incentive to influence the result — don’t rely on an unverified flip of any kind, physical or digital. A flip performed by a party with a stake in the outcome, with no commitment scheme and no witness, is trivially manipulable regardless of how fair the underlying mechanism is: the failure mode isn’t the coin, it’s the missing trust boundary around who controls it. Use a formal commit-reveal protocol, an audited random source, or a neutral third party instead.

FAQ

Why did I just get 4 heads in a row — is the flipper broken? No. The probability of 4 heads in a row under a fair coin is 1/16 (about 6.25%), which will happen fairly often across many flips. Streaks are an expected feature of independent random events, not evidence of bias.

How many flips do I need before I can trust the split looks fair? No hard cutoff, but a few hundred flips will show a proportion within a couple of points of 50%, and a few thousand tighten that further. Ten or twenty flips tell you almost nothing about long-run fairness.

Is a seeded coin flip less random than an unseeded one? Not statistically — both produce a genuine 50/50 distribution over many trials. A seeded flip is reproducible if you know the seed, useful for verification, and irrelevant to fairness as long as the seed was chosen fairly and not adjusted after seeing an unfavorable result.

Does the 51% same-side physical bias mean real coins are “rigged”? No — it’s a small, consistent effect from the mechanics of tossing and catching a coin, not manipulation, and it only shows up across large numbers of tracked tosses. For any single flip it’s not something you’d notice or need to worry about.