A completely different approach might be to gunzip the .graffle file, read its XML text directly, and transform it with XSLT, or simply mine it with a scripting language like Perl or Python. This would obviously bypass the problem of manual ungrouping ...