Within the constraints based on the top-level types which VMD can parse, the libraries gives the end-user the ability to design macros with dynamic data types. By this I mean that a macro could be designed to handle different data types based on some documented agreement of different combinations of macro input meaning slightly different things. Add to this the ability to design such macros with variadic parameters and we have a preprocessor system of macro creation which to a lesser extent rivals the DSELS of template metaprogramming. Of course the preprocessor is not nearly as flexible as C++ templates, but still the sort of preprocessor metaprogramming one could do with VMD, and the underlying Boost PP, in creating flexible macros which can handle different combinations of data types is very interesting.
Of course macros need to be usable by an end-user so the syntactical ability of sequences to represent different types of input data must be balanced against ease of use and understanding when using a macro. But because certain sequences can mimic C++ function calls to some extent it is possible to represent macros as a language closer to C++ with VMD.
What is important when designing a macro in which you parse input to decide which type of data the invoker is passing to your macro is that you are aware of the constraints when parsing a data type. As an example if you design a macro where some input can either be a number, an identifier, or some other data type top-level input then attempting to parse the data to see if it is a number or identifier could fail with a preprocessor error and nullify your design if the data is not a VMD data type. So designing a macro with data types in mind often means restricting data to parseable top-level types.