Abstract: The term surrogation is used instead of approximation, estimation etc, to separate rol playing from how well a role is Played. To obtain a surrogate for an actual but unknown object one needs an algorithm to compute it from input data giving what is known about that object. Invariance arguments are used to show that surrogates must be fixed points of certain mappings associated with the mathematical structure characterizing practical context. The corresponding surrogation algorithms are said to be macroinvariant. Some special cases where there is essentially only one such algorithm are then considered. These results are derived without reference to proximity to actuality, that is to how good the surrogate is. Finally it is shown that if one does not use macroinvariant surrogation, then he could, in principle, do better, in any given metric sense, by means of such an invariant procedure.