Usually, the behavior of the unparser is symmetric to the
behavior of the parser; however, there are cases where the DFDL schema will
accept several equivalent representations for the same logical data. In this
case it would be ambiguous which of these equivalent representations should be
produced by the unparser. The DFDL standard contains representation properties
which are used to eliminate this ambiguity. It is a schema definition error if
a DFDL schema is being used to unparse data and there is any ambiguity about
the representation.
We can either ignore such issues, because the context doesn't require us to consider this a requirement statement, or reword so as to avoid RFC2119 terms. I am not sure it is worth changing this prose in all the places where this sort of thing happens. The word 'should' is less problematic than the word 'must' of which there are hundreds of occurrences.
Thoughts?