"Heads or tails" is the default stand-in for a perfectly fair 50/50 outcome. It's used in math class to teach probability, in sports to decide who kicks off, and in arguments to settle a tie. But a physical coin flip isn't actually a coin toss in the idealized sense -- it's a spinning object obeying ordinary mechanics, and mechanics has an opinion about which side lands up.
The wobble that breaks the symmetry
When you flip a coin with your thumb, it doesn't just rotate cleanly around its central axis. It also wobbles -- precesses, in physics terms -- around a second axis, the way a spinning top tilts as it slows down. Mathematician Persi Diaconis and his collaborators modeled this in the mid-2000s and showed that the wobble means a coin spends slightly more of its airtime with its starting face pointing up than pointing down. The coin isn't "remembering" which side started up through any mystical means -- it's a direct consequence of how an off-axis spin behaves, and it holds for a normal human flip, not a perfectly controlled mechanical one.
Their prediction was specific: a flipped coin should land on the same side it started on about 51% of the time, rather than the 50% you'd expect from a fair coin.
Someone actually tested it on 350,000+ flips
A prediction from theoretical mechanics is one thing; confirming it requires a lot of actual coin flips, recorded carefully enough to know which side was up before each toss. In 2023, a large team of researchers organized exactly that -- volunteers filmed themselves flipping coins (dozens of different coin types and currencies) and logged the starting side and the result for each one. The dataset ended up with over 350,000 individual flips.
The result lined up with the decades-old prediction almost exactly: coins landed on their starting side about 50.8% of the time. No bias toward heads over tails specifically -- the bias is toward whichever side was up when the flip began, which is why it never showed up in simpler studies that only tracked heads-vs-tails counts without recording the starting position.
Why a less-than-1% edge is still interesting (and still irrelevant to you)
For a single flip to decide who gets the last slice of pizza, a 50.8% vs 49.2% split changes nothing you'd notice. You'd need thousands of flips, all starting from a known side, before the edge would reliably show up at all -- and almost nobody tracks or exploits this in casual use.
Where it matters is anywhere a tiny statistical edge compounds over a huge number of repetitions: casino-style games, large-scale randomization in research, or any process that assumes a coin flip is a perfectly unbiased 50/50 primitive. If you're building something that depends on true fairness at scale -- a lottery draw, a randomized study assignment, a tiebreaker used thousands of times -- a physical coin is quietly the wrong tool, even though it feels like the most "random" option available.
The fix is to stop using a physical coin
The bias only exists because a real coin has a starting orientation, a center of mass, and a wobbling spin that a human thumb imparts. Remove the physical coin and the whole mechanism disappears. A random number generator doesn't have a side that started up, so it has no wobble to exploit -- it just needs to pick between two outcomes with equal probability, which is a much easier problem than physics.
That's the case for flipping digitally when fairness actually matters, even for something as low-stakes as a game night decision: you're not trading a "real" flip for a fake one, you're trading a measurably biased physical process for one that isn't.
If you want a flip without the wobble, the Coin Flipper gives you heads or tails with no physical bias to account for -- useful for a quick decision, settling an argument, or anywhere you'd rather not think about which side was facing up before you started.