[Quality] another can of worms

[Ed Shaya]

Martin Hill wrote:

> Hi all
>
> I've been wondering what to do about quality, and how to describe it.
>
> We have a general one - if we allow anyone to publish to the VO, how 
> do we describe a datacenter's quality?  There are datacenters setup 
> for a sky survey, with all the processing involved that has been set 
> up by teams of astronomers (presumably?) double checking each other. 
> And there is data published by a small group or an individual so that 
> people can access it, but which has not gone through such a rigorous 
> set of tests.  We don't really want a general query to the registry 
> returning both as 'equals value' (or do we?).

I share your worry about inaccurate in error bars slipping in. But, if 
we try to fully answer this question we will be embroiled in politics 
and the NSF has funded the NVO project with the hope that it would not 
be 100% politics.  We should focus on more technical issues.  The most 
expedient thing to do is to allow all to publish and trust error bars.  
Creating a system that can properly handle data with error bars is both 
novel and comlicated enough. 

Query responses should carry information on the origin of any data 
unless explicitly told not to.  Then as a separate effort (perhaps IAU 
backed)  examine rational ways to censure data.  We should keep in mind 
when developing metadata standards that we have sufficient information 
to create automated means of hunting for an eliminating poor data or 
poor quality determinations.

>
> We also have a smaller scale one - each item of data may need to be 
> marked.  For example, a sky survey may have items that have been 
> marked as 'possibly satellite track' or 'instrument feature' (is that 
> the word? such as diffraction spikes?)
>
One can provide for a Note to be added at any level down to the 
individual pixel.  A single Note can be referenced many times for reuse 
in a document.  A Note can carry  a pixel list that indicates where in 
the data it applies.   This concept should be carried into the Data 
Model in general.  Notes can hold any string but it could be mixed with 
elements.  The Metadata group could perhaps work on some standard 
elements for common notes like <cosmic ray hit/>, <cirrus/>, 
<telescopeBumbedtheConsole/> or <diffraction spike />

>
>