ACID, roundtrip and Odf generation

1) ACID should be fairly simple to implement. Perhaps like the tests
here http://testsuite.opendocumentfellowship.com/ with the result
here.
2) Roundtrip tests should be more concerned with how applications
handles unimplemented elements (they should be kept intact). This
might be tested with documents that are just opened, resaved and
diffed.
3) But how are the odf generation parts of the applications to be
tested? I.e. Should automation tools be used? Using the GUI for each
application? How could the resulting xml be normalized to be
comparable?

I think application interoperability depends on all three parts.

/$