Well for visualizing it, that's what the keyed dependency graph in the Dependency Editor is for. You can see exactly what it's doing over the full range of values. It's very easy to spot outliers. Cubic I guess, whatever it is I've always felt that linear interpolation would be much friendlier.