Heads or Tails: What a Fair Coin Flip Actually Looks Like
Real coins land same-side-up 50.8% of the time, streaks of six are normal, and a browser can flip fairer than your thumb. Here's the evidence.
A real coin flip is not quite fair. In 2023, a team led by František Bartoš recorded 350,757 flips by 48 people and found the coin landed on the same side it started on 50.8% of the time. That confirmed a prediction Persi Diaconis made back in 2007: a tossed coin precesses — wobbles around its axis — and spends slightly more time with the starting side up. Small edge, but real. If you always start heads-up and your friend always calls tails, you’re skimming half a percent off them.
A virtual coin flip doesn’t have this problem. There’s no starting side, no thumb technique, no precession. One random bit decides the outcome, and both values of that bit are exactly equally likely.
Where the randomness comes from
Browsers ship two random generators, and they are not interchangeable. Math.random() is a pseudorandom generator seeded once — fast, fine for shuffling a playlist, and predictable if you know its internal state. crypto.getRandomValues() pulls from the operating system’s entropy pool, the same source used to generate encryption keys. That’s the one this tool uses. Nobody can predict the next flip, including the code doing the flipping.
Does it matter for deciding who does the dishes? Honestly, no. But “cryptographically fair” costs nothing here, and it means the tool holds up even when something real rides on the result.
Streaks are what fairness looks like
Flip 100 times and you’ll probably see six-in-a-row at some point — the odds of at least one such run are around 80%. People find this suspicious. It’s the opposite: a history with no streaks would be the rigged-looking one, because genuine randomness clumps.
The famous cautionary tale is the Casino de Monte-Carlo, August 1913, when black came up 26 times running at roulette. Gamblers ruined themselves betting on red, certain the wheel owed them a correction. It didn’t. The wheel has no memory, and neither does a coin — after five heads, the sixth flip is still 50/50. The streak counter in the tool exists partly so you can watch this play out yourself.
Why the percentages take forever to settle
The stats panel shows heads and tails percentages, and if you flip ten times you might see 70/30. Nothing is wrong. Percentages converge slowly: after 1,000 flips the split typically sits within a couple of points of 50/50, but the absolute gap between heads and tails counts actually tends to grow. The law of large numbers promises the ratio settles, not the difference. That distinction trips up almost everyone the first time.
So use the sample size that fits the decision. One flip settles who takes out the trash. If you’re testing whether the tool is fair, give it a few hundred and watch the percentages drift home.
Older than the euro, older than the denarius
Romans played this game as navia aut caput — ship or head, after the designs on their coins. The stakes are occasionally bigger than dishes: every NFL overtime starts with a toss, and in 1969 the Milwaukee Bucks won a coin flip against the Phoenix Suns for the first pick of the NBA draft. They chose Lew Alcindor, later Kareem Abdul-Jabbar, and won a championship two seasons on. Phoenix, who lost the flip, has never won one. That’s a lot of franchise history hanging on a single bit of information.
When you need a fast, unarguable 50/50, flip the virtual coin and let the entropy pool decide.