I guess that's the source of Visual6502's discrepancies as well. I might be wrong here but AFAIK it doesn't try to model the analog behaviour at all. Does ANE give any of the $ee stuff in it?
Just tested page crossing with SHX/SHY on a real machine, and the page indeed changes to the value stored. So you can drop the "and may even not exist at all" on page 38.
Re: "and may even not exist at all" - seems to me it's an issue of contradiction rather than a case of missing proof. :)