For us, dividing fields between Work, Instance an HeldItem has proved to be rather easy. The BIBFRAME concept is applied in developing the datamodel for NBC, the new discovery platform for the Dutch Public Library System.
In collaboration with OCLC developers in Leiden, The Netherlands, we created FRBR-expression-like authority records in GGC, our shared cataloging system, that have lead to near-perfect clustering in front-end implementations.
See  for an explication of our two-level FRBR implementation and  for the Aquabrowser-implementation at the Public Library of Amsterdam.
Cheers from the Netherlands,
Op 11 nov. 2014 om 18:21 heeft Robert Sanderson <[log in to unmask]
> het volgende geschreven:
My interpretation (which does not imply endorsement) is that the majority of bf predicates are currently limited by domain and range of either Work or Instance, and the intent is that Work / Instance / HeldItem are intended (but not formally specified) to be disjoint. There have been many discussions on the list and elsewhere about whether a predicate is to be used on a Work or Instance, and how hard it is to generate that from MARC records. The resolution of that complexity is non-trivial and inevitably there will have to be trade-offs between purity and practicality.