So Bibframe decided a "top-down" approach, where items and holdings in a library are just descriptive assets to a Bibframe abstract work or a concrete instance, residing on the same level as cover images, comments, reviews, tags etc. from third parties.
One challenge is, library holdings do not change once they are created, they are of persistant nature during the lifetime of an item, whereas other annotations are quite volatile or even obsolete in the perspective of catalog building.
A logical challenge is the nature of library items. One of the most important tasks of cataloging is making items and holdings information persistent and future-proof. Within the Bibframe approach, there is an inherent mismatch of the semantics of holdings and other annotations.
The library which creates item and holding information should get included into the model itself. It is important to contextualize the valuable cataloging process itself on the Semantic Web.
Instead, Bibframe decided that only information about the Work dominates in the future perspective on catalog data modeling. In other words, without having Works/Instances in a model, there is nothing in a library.
On the other hand, a "bottom-up" approach would honor the nature of
library items better. That is, starting from information about an item in a catalog, the cataloging institution is authorizing the item identifier (mostly a barcode or a shelf mark). From acquisistion, there are also many external identifiers at this level.
Derived from the item, physical information could be collected, which are common facts that are true across all cataloging rules in the past, present, and future. How does the item look like, is it hardcover, softcover, what extent, what service does a library offer for this item? How can the item be transported to the patron, or between libraries? What about the Web of Things? RFID?
Next, the publisher information could be extracted. For instance, the name, place, and date of publication. But, these are second-class information sets, as publishers have control over this kind of information, not the libraries. This information could propagate to a higher level of description, for sharing this information between libraries, in union catalogs, on the Semantic Web, wherever. This would save repetitive work. This holds also for any other third-party, non-publisher-made information set, that could be shared. And on this level, I would like to see tags, annotations, cover art, comments etc.
This process of catalog building could be continued over the sources of information along the provenance of an item.
In the end, there is the title information as highest abstract entity, in the form the creator or editor gave it to the work. Of course, the information set hierarchy could be organized in the many aspects and relations we know from FRBR, or by other more refined rules.
The information sets that are created in such a bottom-up process could be matched automatically level-by-level, in order to build union catalogs with consistent and less redundant information.
In the Bibframe model, the order of items and holdings and the best method to add private or public annotations while avoiding redundant work is not totally clear to me.
My hope is also, by using a library-centered bottom-up approach, many librarians could be motivated to participate in the process of migrating into a new format.