Understanding linked data fragments

Hi all,

I'm working on a project in which JSON-LD and Hydra have been selected as
key aspects of the project's architecture and implementation. I've been
tasked with researching a number of areas to better understand how JSON-LD
and Hydra are to be correctly made use of, and a picture is slowly starting
to form regarding how all of this fits together. One thing that is
currently unclear to me, despite having read the specs multiple times, is
exactly what problem "Linked Data Fragments" are solving, other than to
simply provide a standardized hypermedia-enabled grouping format for a
subset of returned data, its metadata and hypermedia controls. Ruben's
publications have talked about LDF as being a middle-ground solution to the
impracticality of analyzing data dumps versus the heavy server load
inflicted by SPARQL queries, but I'm having trouble deriving from the
specification the actual mechanism by which LDF provides a querying
solution. I imagine that if LDF were merely describing a unified
metadata/hypermedia/data response format, then there would not be talk of
it providing a querying solution, as this would be a problem left up to the
server software implementor.

Is the idea to provide cacheable preconstructed responses to common queries
using the LDF format, with the hypermedia links to other LDF documents,
thus providing the mechanism by which the entire query result can be
iterated over? I have a feeling that that is not it at all. So then what am
I missing? If I may offer feedback on the specification, a picture speaks a
thousand words. Some kind of diagram, or simple example architecture
illustrating how LDF is supposed to work from a client-server perspective,
would go a long way toward facilitating a better understanding of the LDF
concept.

If my question is too large, I'm not out to waste anybody's time - even a
pointer to an article, blog post or other publication that answers my
question would be appreciated.

thanks,
Nathan Ridley