Here are a few suggestions:
XSLT 2.0 has been a recommendation for several years and (more or less)
soon, we will work with XSLT 2.1 and XSLT 3.0, too.
On the other hand, many implementations only work with XSLT 1.0 or have
only partially implemented (the now retired XSLT 1.1 or) XSLT 2.0.
This is not seldom challenging when it comes to type strength issues, for
example. Writing valid XSLT 1.0 that is also easily portable to XSLT 2.0
is sometimes a bit problematic ...
Focusing on versioning issues and how to overcome them would be great! For
example when it comes to sorting and grouping (of course, I'm thinking of
Muench, etc.) in different XSLT versions especially when working in
multilingual environments using different parts of Unicode (not only the
BMP - Basic Multilingual Plane - but also other planes, where several
implementations are flawed), with different alphabets and collating
algorithms, ... Which work/walk-arounds should be used?
How to take advantage of schema awareness within XSLT and how to
work/walk-around it when it is not implemented is also something I would
really like to know better.