Error column in VOResource

[Roy Williams]

Clive

> > But what happens when there is an error ellipse, RA, Dec, and the three
> > components RA.RA and Dec.Dec and RA.Dec of the covariance.

> How often does this happen Roy?

Of course this is the right question. If it is bizarre, then let's forget
it. But if it is a small part of a much larger whole, we don't want to paint
ourselves into a corner.

(1) Some technology ideas

In RDF, we have triples: "Subject Relationship Object". I would say that
"x_err    is-error-of    x", where "is-error-of" is a well-defined
relationship.

Similarly we could say that "bad-x-data    is-flag    x", meaning that a
column of flags referrs to another column.

In a wider context, we could say "RA    is-member-of    Equatorial-coord"
and similarly with Dec. We can define a covariance matrix Coord_err in terms
of its constituent members. Then say that "Coord_err    is-error-of
Equatorial-coord".

We could define equivalence between Equatorial-coord and Galactic-coord
objects. We could use RDF say that collective objects like these
can-be-handled-by specific software packages. We could do the GROUP concept
with RDF. The UCD structure can be subsumed into this RDF technology. You
can say anything with "Subject Relationship Object"!!

(2) Pragmatism

Now we get a whiff of the UCD3 and ontology questions that we discussed last
year. Who will create and build all this fancy metadata, who will write the
software that reads it, and who are the clients that want to use it.

So maybe I come round to agree with you. The most common "relationship"
between table columns is one of univariate error estimate. So lets just get
that right and forget the rest. But if that is the philosophy, we should
forget quite a bit of other VO-related "modelling" activities, right? Maybe
all this stuff about "Quantity" and "Observation" is just the PARAM element
in VOTable?