[gtr]

I enumerated the sequences that the article mentioned, and counted how often a tail filled a head, and vice versa, and got 12 instances where a head follows a head, and 10 where a tail followed a head. So there is a difference for just counting up all the possible 4 flip sequences where at least one of the first three is a head. However, doing a randomised test where I generated a random 4 length sequence, rejecting it if none of the first three was a head, then doing the same test showed no real difference.

Code here https://github.com/gregryork/Flips/tree/master/src/flips

[Beltiras]

Take a look at the updated gist. There's an asymptote for P(H|H) when trials grow approaching p.

EDIT: Graph is empirical H|H against run length (output of the last function in a linear graph)

[gtr]

I'm not really sure what I am looking at there. Is the X axis the number of trials run, or the length of the run?