You'd think that should be the case, and given 2119 it is all that
makes
sense, but there are way too many cases where the subject turns out to
be (explicitly or implicitly) "authors of future RFCs".
In RFC 2542 ("Terminology and Goals for Internet Fax") I wound up
using a notation {1} {2} {3} to replace MUST/SHOULD/MAY when talking
about what future RFCs should do
{1} there is general agreement that this is a critical
characteristic of any definition of ....
{2} most believe that this is an important characteristic
of .....
{3} there is general belief that this is a useful feature
of ....., but that other factors might override;
a definition that does not provide this element is
acceptable.