Multiple Contexts with expand/compact?

Dear JSON-LD folk,
  (and cc iiif-discuss)

We're trying to provide useful contexts, frames and sample code for working
with the JSON-LD representations of our data in the IIIF community (
http://iiif.io/).

One issue that we're struggling with is how to provide use contexts across
multiple, related specifications.
There are two APIs in IIIF. One describes the structure of a set of images,
and another describes how to interact with the individual images.  The
intent is that the two can be implemented separately or together -- the
structure could reference static images, and the image service might be
stand alone without any structure overlaid.

>From the structural description's JSON-LD, we import the context of the
image description only at the block where it's needed.

A completely artificial example is at:
    http://iiif.io/api/presentation/2.0/example/test_case.json

When the document is expanded, the @context in the embedded service block
is processed, and the correct RDF representation is generated.
However, the @context for the block is lost, so when it's later compacted,
the remote fields don't get returned to the nice JSON keys they started out
with.

Is there a solution for this, other than having a single context with all
of the mappings in it?  This doesn't help for external contexts however, as
we expect to make use of the GeoJSON-LD context and structure.  For more
about our approach in this regard, see:  http://iiif.io/api/annex/services/

We have also been using framing, which runs into the same problem, as it
expands the document in order to frame it.

A real example is:
    http://iiif.io/api/presentation/2.0/example/manifest1.json

Many thanks for any thoughts on the matter,

Rob Sanderson

-- 
Rob Sanderson
Technology Collaboration Facilitator
Digital Library Systems and Services
Stanford, CA 94305