The XPathEvaluator_evaluate_TYPE_ERR selects the document root and asks for a
number result. This is a completely valid conversion in XPath and shouldn't
raise the TYPE_ERR exception. The number function converts a nodeset to a string
as if by a call to the string function and then converts the resulting string.
The string function converts a nodeset to the string-value of the node in the
node-set that is first in document order.
See http://www.w3.org/TR/xpath#function-number and
http://www.w3.org/TR/xpath#function-string

One way to get a TYPE_ERR is for example trying to get a node iterator or a
snapshot or a node from an expression that returns a string or a number or a
boolean (eg. "string(something)" and evaluate that to FIRST_ORDERED_NODE_TYPE).
The inverse doesn't seem possible since the string(), boolean and number()
functions defined in http://www.w3.org/TR/xpath#corelib are used to convert the
result to the requested type for STRING_TYPE, NUMBER_TYPE and BOOLEAN_TYPE. As
you can see those conversions are defined for converting from all other result
types.