In general, in all Schemedoc endeavors I've been involved with, I've
tried to use the "funnel" pattern.
Ideally there's a "single point of truth" which is located as close as
possible to the authoritative source of that information. For example,
since a Scheme implementation is the authoritative source on its SRFI
support, having the point of truth in the implementation is most direct.
That source information is then fed through a filter, whose output may
be fed through another filter, etc. to produce whatever useful outputs
people want. The source information is quite delicate because it must be
written by hand. The filters are not similarly delicate because they are
code that is self-documenting, can be fixed, and ideally doesn't need to
change much over time.
A key constraint of the funnel pattern is that information flows one way
only. There are no loops; this means you can build a directed graph of
the information flowing downstream. Ideally we could aggregate
everything into one place at api.scheme.org using this pattern.
