As decided at meeting 147, a sentence has been added (in the editor's draft) to section 1.1:
    In the absence of an implementation-defined way to
    differentiate, all markup creates token boundaries.

However, XML's definition of "markup"
    <http://www.w3.org/TR/2006/REC-xml-20060816/#syntax>
is perhaps broader than what we had in mind when we said "all markup". For instance, it seems unlikely that we meant for a character reference
to create a token boundary. Similarly for entity references and perhaps CDATA section delimiters.

We could be more specific about which kinds of markup we mean, but instead, maybe we shouldn't be relying on the idea of markup. Full-Text operates on instances of the XQuery/XPath Data Model, where markup doesn't exist. So, for example, we might say:
    In the absence of an implementation-defined indication otherwise,
    a token must not contain characters from more than one node.
(although we might have to make that more precise).