I'm pretty sure the algorithm has not been wholly (publicly) specified yet. Much of the hash tree structure is, as are parts of the compression function, but not the whole thing. Probably it will not be fully described until later this fall, closer to the NIST deadline.