I and Alberto Ferrari recently completed the writing of The Definitive Guide to DAX, and we spent months to correctly describe the internals of evaluation context in this language. There are many details that make data model working with both DAX and MDX, and sometime there are behaviors that are not intuitive to understand.
A function that seems to work like magic is ALLSELECTED, which is very useful when you create measures that will be used in Excel pivot tables. What is not obvious is that the DAX engine has to realize what the user is selecting on a pivot table that generates a query in MDX. In reality, there is no other communication between client and server other than the MDX query, and ALLSELECTED is not related to MDX, it is a DAX function!
Alberto extracted from the book part of this description and published the Understanding ALLSELECTED article on SQLBI. You will see that the magic in this function is just a particular manipulation of the filter context, which keeps track of the iterated table in the filter context every time a context transition happens. Not clear enough? Well, the article explains this better!
Returns all the rows in a table, or all the values in a column, ignoring any filters that might have been applied inside the query, but keeping filters that come from outside.
ALLSELECTED ( [<TableNameOrColumnName>] [, <ColumnName> [, <ColumnName> [, … ] ] ] )