Sadly, I think we will not 'get it right' no matter what we do, but  
we should be very careful.

1. The 'test-the-tester' approach has to be flawed in exactly the  
same way as the simple 'test-the-resource' approach is flawed -  
either way it is just as easy to cheat.

2. We know from experience that people decide for themselves who and  
what to trust. It is very important to be able to say by whom, using  
what tests, and when something was tested.

3. It is also important to say exactly what was tested - the  
components of the resource, the resource as a composite object, this  
or that version or, if you like, which instantiation of the resource,  
etc..

4. We have enough research to show that testing does not and cannot  
guarantee accessibility to everyone.

and, what is compelling for me,

5. any individual user only wants to know if the resource will be  
accessible to them at the time of delivery.

We have recommended the AccessForAll approach in favour of  
generalised testing for certification because that way, the  
description of the resource is available for decision-making.

This list, beyond all others, should have many on it who understand  
why we use metadata, and they should be able to contribute to how and  
why metadata is useful in this context. The credibility of the  
metadata is important but so is the content. Specific details are  
often needed for accessibility decision-making. If there are too few  
details, people who could use resources will miss out on them and if  
there is too much demanded, nobody will bother to supply it.  These  
are the sorts of problems that have been at the heart of our work for  
the last few years.

I'd like to think that we can do, as we have always tried to do,  and  
combine the use of metadata to convey information ABOUT the resource,  
basing the statements that make up that metadata on the tests that  
W3C so carefully develops. In my humble opinion, this does not lead  
to a 'let's certify' approach so much as a 'let's describe' approach.

Surely we want to empower the user to make the important decisions  
about what they can/will tolerate just as, when we choose a hard-to- 
get-into restaurant, we let our colleagues in wheel-chairs decide to  
join us or not: it should be their decision, not ours.

Liddy