scatteredinterpolant will ALWAYS reproduce the data exactly, although it may sometimes introduce tiny noise on the order of eps, just due to floating point arithmetic. Interpolation is interpolation.
Others have suggested extrapolation. That has NOTHING to do with interpolation, and prediction of the original points in your set.
Most likely, this is due to a bug in your code. That essentially, you are not plotting the real data that was used to interpolate the surface. That happens.
So, if you want a valid answer, then you need to provide the data. Show how you tried to plot it. Show what you did. Give an example where you think scatteredinterpolant fails. When you do that, along with the data itself, I'll be able to show what you did wrong.