There are several things I would like to try in the future on the Marmousi example. Perhaps I didn't smooth the velocity model enough? How is the smoothing and the ability to interpolate related to the ray validity conditions? Perhaps a 500 m source spacing was too much to ask for, perhaps a 250 m would have be fine? Perhaps solving an overdetermined least squares system (my examples were exactly determined systems) might help to compensate for traveltime gradient irregularity? Is there an intelligent method to smooth or (median?) filter the traveltime gradient fields (input and interpolated) before spatial integration? Finally, is there a better approach which interpolates for the traveltime directly, instead of first finding traveltime gradients and then integrating? If there were, then that would minimize the penalty of having a few interpolation errors being smeared throughout large portions of the final interpolated traveltime result.