Echoes arriving late are weaker than echoes arriving early. Thus data is ordinarily scaled for plotting using some time-variable scale. Should migration be done before or after this scaling? The results will differ in an interesting way. The top part of the hyperbola has flat dip, whereas the asymptotes, which come later, have steep dip. So, amplification of late information coincidentally amplifies the steep dips. I think the main effect of choosing to do migration before or after scaling is selection of the dip spectrum in the final display. A pedantically correct approach is to migrate first and scale second, but the result will be weaker in dip and fault information than the answer obtained by scaling first and migrating second. A side benefit of the latter method is that you can save computer memory by storing scaled values as short integers. I used 16 bit integer storage in my pioneering work. Computations and local storage used 32 bit floating point arithmetic. I see little justification for 32 bit storage generally used today. We can't interpolate between channels to 4 bits of accuracy.