The tool uses the "high-fidelity" Atom output mode for getting at item bodies. That preserves namespaced XML elements and other extra data from the feed. It uses JSON for everything else, and will fall back to regular Atom output if the high fidelity mode is not well-formed (it was added in late 2010, as things were winding down, and thus never got a lot of testing).
Comments
I wrote most of Reader's JSON API in 2006-2007 :)
The tool uses the "high-fidelity" Atom output mode for getting at item bodies. That preserves namespaced XML elements and other extra data from the feed. It uses JSON for everything else, and will fall back to regular Atom output if the high fidelity mode is not well-formed (it was added in late 2010, as things were winding down, and thus never got a lot of testing).