Roll a die until a 6 appears. The first 6 comes on roll 1 with probability 61. To come first on roll 2, roll 1 must fail and roll 2 succeed: 65⋅61=365. On roll 3 it is (65)261. The pattern is P(X=k)=(1−p)k−1p: k−1 failures in a row, then one success. Each extra roll multiplies by 65, so the probabilities shrink steadily but never reach zero.
The single most likely outcome is always k=1, yet the average wait is E(X)=p1=6 rolls. Both are true because the tail is long: waits of 10, 15 or 20 rolls are uncommon but possible, and they pull the average up. The chance of waiting more than k rolls has a neat form, (1−p)k, because it means the first k rolls all fail; for 10 rolls that is (65)10≈0.16.
The die has no memory. After 5 failures, the chance of a 6 on the next roll is still 61, and the expected remaining wait is still 6 rolls. Believing a 6 is "due" is the gambler's fallacy.
The geometric and binomial distributions both use p and 1−p but ask different questions. Binomial: in a fixed 10 rolls, how many 6s? Geometric: how many rolls until the first 6? A fixed number of trials means binomial; stopping at the first success means geometric.
Think of it like this. Waiting for a geometric success resembles waiting for a bus that arrives completely at random, each minute carrying the same chance. Having already waited ten minutes doesn't make the next minute any luckier, so the wait so far says nothing about the wait to come. Real buses run on timetables, so the comparison holds only for truly random arrivals.
Core idea. The geometric distribution gives the probability that the first success occurs on trial k, when independent trials each succeed with probability p.