To assume macros as if they were reductions, is makruh (unacceptable, untrustable). If that macro is not actually with a well-form (VD78), the test is not valid in testing what the full (uncompressed) net would tell.
The issue is about assuming a macro-compressed net, as if that were equivalent to the full. Otherwise, if the net-architect has chosen to manually redesign the net, that is another story. Strategies differ.
To compress a formal-net with macros, is similar to compressing html/text documents. That compressed file is not processable (by an html/text editor/viewer) until restored to its original format.
When going to a hospital, any people who may show the way to that hospital, is good for information. We may not assume, they harbor the medical know-how, though. We follow that path, to get to the (medical) function. A macro is similar. When testing the full net, if the full content of a (sub)macro is not known, the test is not valid.
In contrast, a (well-formed) reduction, is the networking of people who know what they do well, and for other function, they could mediate. For example, there is a well-wrapped remedy (or, function), that your family doctor may refer at (cf. infirmazing). A pill-handout, may not actually suggest that your doctor has made that, nor he/she would need to test that pill in his/her laboratory. He/she is supposed to know how to get to the standard sources of information (& medical supplies), and utilize that in his/her framework, working with people in town.