On another thread, Kevin Ford said
"... the challenge is clearly identifying Instances of Works. Different publishers/places of publication/dates is one of those 'split' points. ISBNs are equally attractive."
Clearly identifying separate Instances is a hard challenge. It is not, however, a new challenge. It is essentially similar to the problem we currently face when trying to decide when to make a new MARC bibliographic record. Unlike Bibframe, MARC is silent on this point, but OCLC has pages and page on this at http://www.oclc.org/bibformats/en/input.html and ALA has a whole book (http://www.ala.org/alcts/resources/org/cat/differences).
Libraries have traditionally been stronger lumpers than Bibframe's proposal to split Instances on ISBNs and have put what has been considered to be essentially the same printing on the same record. Lumping does a few things:
*It reduces the proliferation of duplicates in databases
*It reduces the number of records that have to be created (in the current system it is easier to reuse an existing record)
*In existing catalogs, near duplicates are usually not helpful to patrons when they are navigating results lists and often make the process of placing holds inefficient (FRBRization helps with this somewhat and future systems may make this less of a problem)
There is also the practical problem of how to know whether the book you're holding in your hand is a match for a particular record (or Instance). When something doesn't seem to fit, it might be that you have a different thing or it might be that the creator of the existing metadata has omitted some information, added some erroneous information (this often happens when a cataloger has "derived" a new record from an existing one, but there are other causes) or has interpreted the information on the item differently.
Sometimes even comparing two books side by side, it can be difficult to tell. I am currently involved in a cleanup project in preparation for an ILS migration. One of our problems is that we have old retrocon records that were matched to OCLC records in an automated reclamation process with the result that we ended up with many cases of two records bearing the same OCLC number. Either the two things are the same and should be on one record or they are different and one of them needs to be matched to a different OCLC record number.
My examples from this project are pre-ISBN era books, but I imagine the problem has not vanished (I am not primarily a book cataloger so don't have the experience to say). We had one case where two books had the exact same outward appearance and bibliographic information (title, publication info, pagination) except that one had "illustrated by X" on the title page. Upon examination, this book had unnumbered pages with illustrations whereas the other one had unnumbered pages with photographs. We put these on two records and I think you would consider them two Instances, but the only difference between the records is that one statement of responsibility. If a cataloger had only the illustrated book and the existing record didn't mention illustrations, it would be hard to know whether the record was describing something different or merely omitted some important data.
On the other hand, we had a case where two books looked visually different, but again had the exact same bibliographic information except that one had "Phoenix Books" on the title page above the publisher's name. There appeared to be no difference in content. We put these on the same record. It seems unlikely that the presence of those two words would influence patrons' preferences. There are also cases where one copy of a book doesn't have an ISBN and a newer printing does, but they aren't otherwise different.
Another problem with ISBN-based Instances is that since libraries have not historically tracked most hardback and paperback editions separately, they won't know which Instance to attach their holdings to. I suppose this could be resolved by libraries' asserting that certain ISBN-based Bibframe Instances are the same for their purposes and linking to that group of Instances.
On the other hand, special collections and rare books cataloger often want to do more splitting than Bibframe. They want not just the ISBN, but the specific printing to be a separate Instance.
Does Bibframe have to legislate boundaries between instances?