This is transferred from the comment made by email here:

https://lists.w3.org/Archives/Member/w3c-xsl-query/2013Jan/0003.html

We say in 2.5.5

derives-from( AT, ET ) returns true if AT is derived from ET by 
restriction or extension, or if ET is a union type of which AT is a 
member type.

If this were the case, then when a function expects union(xs:string, 
xs:decimal), it would be a type error to supply an xs:integer. 
Furthermore, "union type" should be "pure union type".

This sentence is immediately followed by a paraphrase that says 
"formally, ...", and the paraphrase gives the a more complete definition. But 
there is no indication that the sentence quoted is to be regarded as 
incomplete or informal.

A better definition might be:

derives-from( AT, ET ) returns true if any of the following conditions 
applies:

* AT is ET
* ET is the base type of AT
* ET is a pure union type of which AT is a member type
* There is a type MT such that derives-from(AT, MT) and derives-from(MT, ET)

The problem is exemplified by test instanceof138. The expected result 
for this test assumes that restrictedDate is substitutable for a union 
type that includes the base type of restrictedDate as one of its members.