Characterization of tree populations resulting from very-extensive heuristic ML searches (10,000 GARLI runs) on four character sets. For each character set there are two side-by-side graphs: A, E for noLRall2 + nt2; B, F for nt12; C, G for nt123; D, H for nt3. The left-hand graph plots the number of searches resulting in a tree with given topological difference score (Y axis), against the topological difference itself (X axis), where "topological difference" for each search result tree defined as the number of nodes collapsed in the strict consensus between that tree and the overall best tree ("best" tree). A dashed vertical line marks the median topological difference for each data set. The right-hand graph plots topological difference from the overall best tree (Y axis) against difference in ln L from the overall best tree (X axis), expressed as a percent of the best score, for all trees with likelihood scores within 0.005% of the best.