Evolution in an NKp genotypic landscape with \(2^8=256\) types. A: The fitness landscape has \(20\) local fitness maxima and many saddles between them, making visualization and evolutionary prediction challenging. Here the landscape is represented as a basin hopping graph (BHG), in which nodes are basins of attractions of fitness maxima and edges adjacency relations between basins weighted by the barrier height. B: As the mutation rate passes a threshold at \(\mu\simeq0.2\) (in units of the maximal fitness difference), the quasispecies distribution delocalizes, as signalled by the inverse participation ratio \(\left(\sum_x Q(x)^2\right)^{-1}/\vert X\vert\). C: The BHG for the effective potential (here for \(\mu = 0.1\)) is much simpler—and immediately predictive, see Fig. \ref{970511}.