As an afterthough, it would be interesting if it is possible to reduce the
number of moves in your fourth phase. The main difference between your
algorithm and Kociemba's is that yours is deterministic. Kociemba's
algorithm performs quite a number of searches before finding the optimal
solution. And even than it is not known whether the solution is indeed
optimal, longer searches might reveal better solutions. Your algorithm
gives an upper bound to the number of moves, and the solution is reached
in limited time. Kociemba's algorithm is in theory unlimited in time.
My experience is that it is best to limit the first phase in Kociemba's
algorithm to 13 moves. But that is only because of time constraints.
dik t. winter, cwi, kruislaan 413, 1098 sj amsterdam, nederland