When using a NodeFilter the XPath for the Nodes (and their parents) in Comparison's details show the "reality" the difference engine has seen, not the one of the document itself. Elements that have been filtered away are not counted which may cause the indices of sibling elements to be smaller than they should be.

See also https://sourceforge.net/p/xmlunit/discussion/73273/thread/92c980ec5b/?limit=25#e143