> There was some opinion (primarily by Matthew Dovey) that
> cql.serverChoice should be used only as a proxy for some other index
> that the client would normally use. The opposing opinion (primarily by
> Rob Sanderson and Mike Taylor) was that cql.serverChoice can do whatever
> it needs to be useful, regardless of other existing indexes.

And I've long held that this is not a meaningful distinction. For example,
suppose  the server has an author index and a title index (but not any
author/title index)  and the server wants server choice to correspond to
author/title.  At first glance, according to Matthew's view, it can't do
that. But all it would need to do is expose an abstract author-title index
(it doesn't have to physically implement that index) and then it can "do
that".

--Ray