Hello,
A response to one of Tom McGlynn's comments:
"Note also, that endorsments mechanism may need to be conditional
"and/or quantitative. E.g., data may be endorsed for astrometric
"purposes but not photometric, or the endorsement may indicate
"that the astrometry is good to 0.1" and the photomtry to 0.1 mag. "
I think the accuracy of the data themselves is a separate issue from the
accuracy of the metadata. This is perhaps another 'policy issue' but the
view which makes most sense to me from previous debates, is that the
responsibility of the Registry is to make sure the metadata describe the
contents of the dataset adequately. That is, there should be metadata to
describe the accuracy of the
astrometry
photometry
time resolution
etc. etc.
Within the catalogue this may be quite complicated (e.g. separate
systematic and random errors, dependence on other quantities) but for the
registry we need to pick out a value or a range for each measure of
accuracy which we decide is important for data-set selection.
If something is missing then the Registry maintenance should make an educated guess wherever possible or cost-effective (some things can be derived from the contents of a catalogue, initially or for major datasets a quick human scan, or eventually links to instrument parameters etc.) - information supplied by the Registry should be noted as such - CDS do similar things such as providing the best coordinates for an object.
If an educated guess is not possible then the relevant accuracy parameter is undefined, and 'undefined' should lead to exclusion of that data set from searches requiring more than a certain accuracy (thus it is important that the Registry does supply at leasts one limit to accuracy where possible).
However the VO should not 'endorse' data on the basis of 'good' or 'bad' astrometry etc. as 'good' and 'bad' depend entirely on what you want the data for.
In some cases the user may specify the required accuracy.
In other cases this may be implicit in the execution of the query, for example calculating proper motions can be done with data of any accuracy which is error-weighted, but there would ahve to be errors recorded or supplied by the VO when a catalogue was ingested. Thus other parts of the VO e.g. supplying algorithms or constructing a workflow would send requests to the Registry and include 'existance of errors in x' or 'error_x < dx' as a selection criterion.
cheers
Anita