Spunk Odds Reference

Coin flip guide

Bamboo Coin Flip: the longest run, on average

Bamboo is a naming choice for this address, not a game variant. Here the question is a single statistic: across a fixed number of tosses, how long is the longest run likely to be?

One statistic, many uses

The longest run in a sequence is a quick test for whether a list of results looks like real chance. A fair sequence of 100 tosses with no run of four or more tails is itself rare: its chance is only 2.73%. A run of six or seven is the typical maximum. Knowing the expected value helps you read your own history without seeing patterns that are not there.

The averages

Expected longest run in n fair tosses (exact)
Tosses nLongest run of tailsLongest run of either face
102.803.66
254.044.98
505.015.98
1005.996.98
2006.987.98

Computed from exact run probabilities: E[L] is the sum over k of P(L >= k).

The longest run grows slowly, roughly with the base-2 logarithm of the number of tosses. Doubling the length of the sequence adds about one toss to the longest run. That is why streaks that look dramatic are routine in long sessions.

Why either face adds about one

A run of either face is a stretch where each toss repeats the previous one. In n tosses there are n - 1 comparisons, and each is a repeat with chance one half. A run of j repeats is a run of j + 1 equal faces. So the longest run of either face is one more than the longest run of repeats among n - 1 comparisons. The third column uses exactly that fact.

How sure is the average?

The average hides a spread. In 100 tosses, the chance that the longest run of tails is at least 5 is 81.01%, at least 8 is 17.02%, and at least 12 is 1.094%. A run of 8 is unusual but not suspicious; a run of 20 in 100 fair tosses (0.003910%) would deserve a closer look at how the results were made.

Applying it to losing runs

Calling a coin correctly is itself a 50/50 event, so the same figures describe the longest losing run. Over 200 calls the expected longest losing run is about 7.0. Any staking plan that cannot survive a run of that length is likely to fail within 200 rounds. Doubling plans are the obvious example, covered on the martingale page.

Is a long run evidence of a rigged coin?

Not by itself. Runs near the expected longest run are normal. Only runs far beyond it, or an imbalance that persists over thousands of tosses, are worth testing.

Does the payout affect the run lengths?

No. Run lengths depend only on the 50% chance of each face. The payout changes what a run costs or earns, not how long it is.