DM Session 3
-------------

Norman Gray presented an interesting suggestion on how one
could extend data models.

- a user can publish a file specifying new utype strings that
are declared to be subclasses of old utype strings.
- other users can then interpret them as just being the old strings
without breaking but users with access to the new model can
make use of the distinctions between them.
- One can declare a namespace interpreted as a URL; the namespace
concatenated with the utype can be interpreted as a URI.
- Norman suggested using RDF syntax to define the subclassing;
JCM suggested one could use XML and have a lower-impact implementation,
since little reasoning is involved. NG said that RDF could also be
generated from such XML.

Mireille Louys, Anita Richards and Francois Bonnarel presented
work on the characterization model.

Anita emphasized usability issues: cf. registry: data providers
will not fill in anything they find hard.

A discussion between Arnold, Doug and Anita about required/optional
axes in STC vs specific DMs like Char. STC was felt to be more
generic and therefore needed to have more things optional.
JCM, DTody emphasized that DMs using Char can add extra 'MUSTs' over
the basic char requirements.
DTody argues that any real data can define all 3 axes.
ARichards says real data providers wont bother

Following on from DM1, there was a discussion about whether
one needed to have 'Not-applicable' as well as 'Unknown'. In
particular, if a theory dataset has no 'time' associated with
it, can one say the time is 'unknown' or must one say that it
is 'not applicable' (NA): "it's not unknown, we know that there
is no time for this dataset".
In Char, absent axes imply axes with properties unknown.
Doug Tody continues to argue for N/A as an option.
SSA needs a way of providing a N/A axis.
Action all to propose ways of doing this, and ensuring
that Char is not in conflict. 

There was a discussion on whether there were problems with
the CharacterizationAxis hierarchy, in particular that child
elements of this object might not have a direct link to the
units hidden in the associated AxisFrame object. Arnold
expressed reservations about this aspect of the Char design
and pointed out that he had come across this problem in STC
and done extensive work to avoid it.

Igor presented a demo which implemented Char in a Russian/CDS
collaboration. Uses XML support in PostgreSQL. Sphere geometry done by
PgSphere later to be replaced by Q3C. Queries using postgres to find
objects with particular properties.
It was pointed out that this was a real life example of using a
data model.

The issue of coordinate systems was raised by Ray: region expressed
in galactic, query in RA, who does the conversion?
Anita: put a tool in the path? Workflow oriented; not the data provider's
problem?

Francois presented a summary of STC use in Char and
a list of implementations.
- Aladin server - VOEvent matching
- Vizier catalog prototype in registry query tool
- (SED constructions)
- HST observations to look for asteroids using SkyBot
Pointed out that SSA is a partial implementation too.

DTody:  Char: what do we use it for?  Some overlap with
SIA already. FB: when we have level3 implemented, useful
for analysis and fine search. JCM: it's a more general and systematic
way of recording the info.

Roadmap: 

Spectrum:  WD by 1 Oct 06 PR by Dec 06 if implementation.
STC:       WD by 10 Oct 06, PR by end 06   
Char:      WD by 1 Oct 06  PR Dec 06
Line:      WD by Oct 06 PR ?