I'm not especially handy. (Get up, those of you who just fell on the floor laughing at the understatement of the century. Stop it. Right now. I know who you are.) Inspired by a recent minor home repair (which would probably be a trivial repair for you normal folks!) involving a screwdriver (flathead, if you were wondering, I'm not quite that inept), I've been thinking about how to explain metadata interoperability in terms of tools.
I've long held that interoperability by prescribing a single way of doing things is unsustainable, even at a relatively small scale, and it seems to me those sets of many-sized gadgets can show us a path forward. Wrenches, bolts, sockets and the like are not all the same. Rather, a bolt is chosen based on what it needs to do - what functions it needs to support and the environment in which it needs to fit. The same is true for descriptive metadata standards. The ones used for a specific class of materials need to both match well with the materials themselves and are supported in the institutional environment.
On the other side, how to deal with a bolt may not be immediately obvious, but it's not all that hard to figure it out. There are many options for what size socket wrench would be needed to tighten or loosen it. It would be nice if the bolt clearly stated what size it was, and this happens in some cases but certainly not in the majority. The wrench needed might be of the type measured in inches or the type measured in millimeters. We can consider these akin to the different approaches to description taken by libraries and archives. A practiced eye can examine the bolt and guess at the right size wrench to try. They'll likely get close, maybe with one or two mis-steps. With trial and error, a novice can find the right wrench as well. I believe the same is true for a system using metadata from an outside source - a skilled human can tell a lot about how best to use it from a quick glance, and trying various tactics out on it will inform how well various choices work for both an expert and a novice. With time and expertise we can transfer some of these evaluations to a system rather than a human (we can do some of this now; for example determining which metadata format from a predefined list is in use can be easily automated), but the process is still the same.
The point as I see it is that for these types of tools there are many options, but each of them are well-defined and well-documented. The same is true for metadata. As we gain more experience, we work towards defining various approaches that work for a given class of materials in a given environment. We then don't need to re-evaluate everything every time we start something new; we can refer to existing knowledge to see that in this case that approach has proven to be useful.
Maybe this analogy fails once you look closer (especially on the tool side - what's here pretty much represents my entire understanding of those types of things), but, like any analogy, it's bound to break down at some point of examination. But I've been pondering a bit and for now I still think it's useful. Feel free to argue with me, though!