Spunk Odds Reference

Coin flip guide

Atomic Coin Flip: variance, spread, and the square-root rule

Atomic is only the tag this URL carries; it describes no special version. The page measures something most guides skip: how widely coin flip results scatter around their average.

Two numbers describe a flip

The expected value tells you where results centre. The standard deviation tells you how far they usually land from that centre. For a one-unit flip paying 1.98x, the net result is either +0.98 or -1.00. The mean of those two outcomes, weighted by their equal chances, is -0.010. The standard deviation is 0.9900 units, almost exactly one unit.

That gap in scale matters. The average loss per flip is a hundredth of a unit, while the typical swing is a full unit. On any single flip the edge is invisible.

The square-root rule

Over n independent flips the expected net grows in proportion to n, but the standard deviation grows only with the square root of n. Multiply the number of flips by 100 and the expected loss rises 100 times, yet the spread rises only 10 times. That is why the edge, small at first, eventually outweighs luck.

Bet of 1 unit(s), win chance 50.00%, total return 1.9800x on a win
RoundsExpected netSD of netRough 95% band (normal approx.)
25-0.254.95-10.15 to 9.65
250-2.5015.65-33.81 to 28.81
2,500-25.0049.50-124.00 to 74.00
25,000-250.00156.53-563.07 to 63.07

Read the last column. After 25 flips the rough 95% band runs from about -10.2 to 9.7 units, so ending ahead is ordinary. After 25,000 flips the expected net is -250 units and the band is about plus or minus 313, so a loss is now very likely.

Where the edge overtakes the spread

A handy crossover is the number of flips at which the expected loss equals one standard deviation. Set n x 0.01 equal to the square root of n x 0.9900 and solve: n is about 9,801 flips. Below that, a typical session can go either way. Well above it, the sign of the result is close to settled.

The chance of being ahead after n flips can be computed exactly with the binomial distribution. With a 1.98x payout you need more than n/1.98 wins to finish above zero. For 101 flips that means at least 52 wins, and the chance is 42.12%. For 1,001 flips it is 37.60%.

What this means in practice

For the general idea of an edge across different game types, see the house edge guide.