Initial thoughts

I have looked over the draft BP - and not seen a lot of comments emerging -
so I'd like to kick of some discussions before I worry about detailed
comments.

The "big ticket" item I feel is that there is a quite extensive list of
Best Practices - at quite a low level - enough to be daunting - but there
is not a clear sense of which set to apply under which circumstances.

Its fair to say that, given the scope is (understandably) proven best
practice - application of these practices is not a guarantee that an
interoperable Web of Data can be built (as current practices have not
achieved such a thing in any substantial way - it is still basically
impossible to discover the nature of content exposed via services).

I think that there needs to be a introductory treatment that a single
implementation architecture is not being specified - and the user is
essentially on their own to choose which BP are applicable to their chosen
approach.

A key question is where relationships are expressed and discoverable - the
links.
There are multiple options, each with pros and cons and some practice - and
these options are not mutually exclusive - but no "best" practice is
identified (or can be yet IMHO). These options include:

1) data publishers are responsible for embedding links into information
resources returned when URIs are dereferenced
2) many resources in arbiitrary locations may embed relationships - and
some crawling infrastructure is to collate these resources and present them
to the user - perhaps by hijacking the dereferencing mechanism to invoke
services to replace or augment the information resources available by
default?
3) Well-known metadata (e.g. robots.txt, VoiD documents in standard
locations for a domain)
4) well-known services where related resources can be registered by a third
party

On top of this are all the questions about canonical formats - RDFa,
JSON-LD, etc - which I think are orthogonal to the content disposition
issue. I think you need to address the disposition contracts before the
usefulness of specific formats can be evaluated.

These are are at the heart of Semantic Web concepts - Open World,
Non-unique Naming and AAA - but there is really no obvious BP for
implementation - in fact most projects seem to deal with these issues by
ignoring them in favour of a centralised repository.

If we are truly going to link data - and especially if we are going to
handle spatial data exposed by specialised service interfaces - then these
basic architectural patterns need to be discussed. Without it I fear the BP
is not going to be easily assimilated or provide useful guidance.

This then has the follow-up advantage of allowing the many BP in the
current discussion to be grouped or cross-references to which patterns they
are relevant for.

Regards
Rob Atkinson