Just to clarify where the S-expression files would come from before the
HTML is finalized -- I'm currently using the class attributes that I
made up to extract things with the Python script. It's quite fast to
mark up a new SRFI so I can generate some metadata quite quickly for
testing/development purposes. I won't be systematically marking up all
the SRFIs in this manner, but I may process a small batch (e.g. 10-20
documents) to see if I can build the web documentation browser / API
from that metadata. This would be a good real-world demonstration that
the metadata format is useful and convenient.

Once the official HTML format is finalized I'll throw away my custom
HTML and update the tool so it gets metadata from the official format.