"One thing I am finding about FRBR (and want to think about more) is
that one seems to come up with different conclusions depending on
whether one works down from Work or works up from Item. The assumption
that an aggregate in a bound volume is an Expression seems to make
sense if you are working up from the Manifestation, but it makes less
sense if you are working down from the Work. If decisions change based
on the direction, then I think we have a real problem!"
The direction one moves in with the WEMI/IMEW model doesn't change the result of using the model in the way you mean. It doesn't invalidates the model or shows a serious problem with the model. It shows that people can often trace complexities of relationships in one direction better than they can in another. So it is a good practice to use the model in both directions when trying to understand it or apply it.
Let's use your case (more or less) as an example.
Consider that we have a published work--Moby Dick--that has additional material published with it, an introduction, a poem, whatever. That extra material published with the work Moby Dick doesn't exist _with_ the text of Moby Dick until we descend down WEMI to the manifestation level. Here's one place confusion enters. What happened to the expression? We just skipped over it. As if it didn't exist. What about the item? We have that in hand. It exists. It's our evidence for the rest. We have analyzed this bibliographic situation from W to I, but we have a gap at expression. That's confusing. There may be something wrong with the model, but before we investigate that, let's look at the situation in reverse, from I to W.
The evidence I have is in my hand. I'm holding the item. It says it is Moby Dick, it has an intro and other stuff with a block of text that looks like it could be Moby Dick. We'll assume for now that it is. From the item, I infer the manifestation.
(Jonathan and others would very sensibly call this manifestation the set of all the more or less interchangeable items, but I prefer to think of it as an _idea_ we have about a set of like items. To me, it doesn't make the best sense to think of the manifestation as a physical entity--a set; it's more useful to think of it as an idea we have about a set of physical entities. This approach keeps the focus on the FRBR manifestation as an abstraction. It avoids reifying the concept, which, I think, is a source of much confusion about the FRBR WEMI model.)
From the FRBR model we know that a manifestation is the embodiment of an expression. From the manifestation, we infer another level of thinking about the item in hand, another abstraction, the FRBR expression. Going up the IMEW ladder, we see there is no gap where the expression should be. The expression is simply an inference we make from the manifestation according to the model. It's a formality. According to the model, an expression for the augmented/supplemented/whatevered Moby Dick exists. It must. And from the expression, let's call it "Moby Dick+a E", we infer the work, "Moby Dick+a W", again, according to the model. So working up the IMEW model, we see the augmented/supplemented/whatevered Moby Dick that I'm calling "Moby Dick+a" is a work, an expression, a manifestation and item.
Coming down the WEMI model, we skipped over the expression level. Why? I think it is because of a couple of things common to how we think. First, when we use the WEMI model in this top-down direction, we tend to reify the abstractions and look for "real" instances of them. Second, when we move down the WEMI model, we deduce the next level from the "evidence" of the one above or evidence from the physical world. Since the abstract levels of the FRBR WEMI model provide no evidence for deduction, and there is no evidence of an expression in the item, and all there is to rely on is the model's claim that "there be expressions here," then we don't see the expression as real. Working up from the item, the step at the expression level is more clear and more clearly a formal part of the modeling process. It isn't a different decision about expression, it is a different view of the model that allows us to more clearly see the expression.
Is this way of thinking, useful? It may be, when or if we think the editorial work that created the augmented/etc. Moby Dick, is worth noting and tracking. Consider for instance the 150 the anniversary edition of Moby Dick published by the Northwestern University Press in 1991. It may make sense and provide some utility for readers for cataloger's to consider this edition a different work than the Norton Critical Edition, 2d edition, of Moby Dick. Because we like to relate a work to a creator of the work when we can, I'll point out the creator of each of these works is the editor or editorial group that edited the text of Moby Dick-if they did that--and compiled the edition. And we might distinguish them by use of the editor's name or the publisher's as we do in this case.
Returning to "Moby Dick+a" for a moment, I want to point out a complexity that I skipped over so far. There is more than one work involved in "Moby Dick+a." The first is the edition itself, "Moby Dick+a," a second is "Moby Dick," itself, a third would be the introduction written for this edition, etc. It would be possible to have the same work/expression of "Moby Dick" in two different "edition-works" of Moby Dick. If the same text of "Moby Dick" is simply repeated in a new context of apparatus--introductions, afterwords, etc., one could have a work/expression "Moby Dick+a" and another "Moby Dick+b" that each contains the same work/expression, "Moby Dick." What makes sense to me is noting and tracking both of these--the edited augmentation and the core work. Other works within the augmented work may also be worth noting, etc., but how far one would follow that path depends on the implementation goals.