Continuing the discussion from Book of carefully-edited songs, with CD: what Works?:
I've been reflecting on this. I think @CallerNo6 has a really important insight with FRBR-Expression .
If you haven't read Functional Requirements for Bibliographic Records by the International Federation of Library Associations and Institutions (IFLA), I strongly recommend it. It has five well-considered concepts, which I find very helpful when thinking about the MusicBrainz data model:
work (section 3.2.1): "a distinct intellectual or artistic creation. A work is an abstract entity; there is no single material object one can point to as the work.…" A MB-Work is said to correspond to an FRBR "work".
expression (section 3.2.2): "…the specific intellectual or artistic form that a work takes each time it is 'realized'.…" A performance of music is an "expression". The choice of notation that goes into a music score is an "expression".
manifestation (section 3.2.3): "the physical embodiment of an expression of a work.… manuscripts, books, …sound recordings, …etc. As an entity, manifestation represents all the physical objects that bear the same characteristics, in respect to both intellectual content and physical form." A MB-Release corresponds to a "manifestation". An edition of a music score is a "manifestation".
item (section 3.2.4): "a single exemplar of a manifestation. …[A] concrete entity." When I enter a MB-Release (manifestation), I have a physical CD in my hand (item). I and MusicBrainz count on my item being representative of the manifestation, and all items from that manifestation having the same recorded music content and cover art.
Aggregate and Component Entities (section 3.3): "The structure of the model, however, permits us to represent aggregate and component entities in the same way as we would represent entities that are viewed as integral units. That is to say that from a logical perspective the entity work, for example, may represent an aggregate of individual works brought together by an editor or compiler in the form of an anthology…."
If the FRBR-Work is the abstract entity corresponding to a creation, that means that performing an FRBR-Work means choosing a specific choice of notes. The set of notes is an FRBR-Expression. An arrangement is a different FRBR-Expression. A composer's later revision of the creation is a different FRBR-Expression. Representing a set of notes in musical notation is an FRBR-Expression. A musical performance is an FRBR-Expression. It is actually an aggregate FRBR-Expression, based on the music score FRBR-Expression, and the edits and corrections that performer makes to the music score (whether or not written down), and the performance choices by that performer, and the implications of choice of instruments and performers.
Committing a set of notes in musical notation to a PDF file or a print run of music score books is an FRBR-Manifestation. Recording a musical performance is an FRBR-Manifestation. The producer and mixing engineer take one FRBR-Manifestation, the original studio masters, and make an audio recording suitable for pressing onto CDs. The resulting master recording is an FRBR-Manifestation. I think you could make an argument that the mixed audio signal, separate from its recording medium, is an FRBR-Expression.
As @CallerNo6 points out, MusicBrainz lacks "an entity that matches the FRBR-Expression". Therefore, "MB-Works sometimes do 'double duty', acting as both" [FRBR-Expression as well as FRBR-Work].
This brings me back to my earlier thread, Book of carefully-edited songs, with CD: what Works?. I think that my difficulties representing that score+CD combination in MusicBrainz stems from the problem that the music score is an FRBR-Expression, and MusicBrainz lacks a data model for handling Expressions. There's no way to express a catch-all Expression for a Work, no way to say that Expression B is related to Expression A, no way to say that an editor made choices that are critical to what makes Expression B important, no way to say that Expression B has a variety of musical information (e.g. translated lyrics) that were not used to make Manifestation C (the instrumental-only CD included with the score book).
If MusicBrainz lacks the data model for FRBR-Expressions, I should not be surprised that I have a hard time expressing them in MusicBrainz. So, I should do the best I can. This includes creating MB-Works to serve as the FRBR-Expression of the edited score. This also includes using annotations to describe what the current MB data model cannot represent.
Thank you to everyone on that thread for clarifying my thoughts. And, thank you for pointing me to the Functional Requirements for Bibliographic Records document.