In this article we revisit a subject that has partly already been examined in previous studies: the behavior of tomographic reconstructors in adaptive optics systems, facing to an atmospheric profile (C2n(h)) different from the one they've been optimized for. We develop a new approach for that. The current usual approach is to simulate the performance of the reconstructor when slightly varying the C2n(h) profile around a nominal one, and show how far the deviation may go. This has the disadvantage that, as the parameter space for potential errors on the C2n(h) profile is basically infinite, it is particularly uneasy to span. Our approach consists in deriving a sort of sensitivity function, that we call vertical error distribution (VED), from the knowledge of any tomographic reconstructor. This function can be computed even for non-tomographic reconstructors, ground-layers reconstructors, single-conjugate AO reconstructors, etc. In any case, it allows us to derive the error when applied to a particular C2n(h) profile, have a direct, global visualization of the error variation with layer altitude, for any number at any altitude. This also allows us to understand what a given reconstructor is sensitive to, at what altitudes or altitude range, or explain why some GLAO reconstructors may perform better than optimized MMSE tomographic reconstructors if low-altitude layers pop up. We also discuss the case of ELTs and apply our approach to large scale reconstructors.