This has been sitting in my Zotero TBR for two years but I finally got to it. If you're a metadata nerd in any discipline, you'll probably find it interesting. Karen Coyle proposes that the FRBR "WEMI" model has accidentally been useful outside of libraries, but to be truly useful needs some revision to make it more flexible.
There are various reasons for not requiring the presence of every entity in the full stack for valid adherence to the model including the lack of omniscience on the part of metadata creators and the potential lack of completeness of a set of metadata at a given position in the data creation workflow.