As I said, a non-linear distribution can easily be fit into a quadratic curve, if logarithmic. Multidimensional curves, ie lots of holes in the data, do make n

As I said, a non-linear distribution can easily be fit into a quadratic curve, if logarithmic. Multidimensional curves, ie lots of holes in the data, do make no sense to model for the initial pivot calc. Binary search finds them much better then. The cost is similar to a hash table, almost constant with 2-4 branches, and mostly into cached data.