The Dangers of Cannibalizing Data Models

[Rob Seaman]

A couple of comments in support of Arnold's self-consistent world view:

> I have always objected to the practice of cherry-picking components from STC, since it destroys the integrity of the model: there is no guarantee that the metadata are either self-consistent or complete anymore.  And Utypes are a form of doing this.


From http://www.ivoa.net/cgi-bin/twiki/bin/view/IVOA/Utypes :

	How can I find which Utype is relevant for my data collection
		• Select an IVOA data model to which you want to be compliant. Ex : Characterisation Data Model
		• Browse the Utype list and its documentation to select the appropriate Utype list for your metadata description
		• Check wether all mandatory Utypes are filled in .

> Pierre le Sidaner complained that in ObsTap he could not characterize time appropriately.  Of course not: it contains a Time Scale, but adding a Time Reference Position was deemed not to be necessary.

Without a VO hermeneutics DM, it isn't clear whether "necessary" is the same as "mandatory" - but one should imply the other.

> Don't blame Pierre; blame the cherry-pickers.  I still believe Utypes are a bad idea, because they only reinforce this practice.

In addition to the more fundamental problem of how the heck do we explain Utypes to civilians, are we clear among ourselves which parts of the various DMs are mandatory and which are not?  (And if they aren't mandatory, are they really necessary?)

> But, as I said, I know this is a lost cause.

As is true of all the best causes…

Rob