Hm, I think this is reasonable, but I'd welcome another CWG opinion on this. For example, are we implying here that "value" should generally be understood to imply lvalue-to-rvalue conversion, and we need to review the standard for other such uses, rather than, say, considering every evaluated expression to constitute a value (of the expression's value category)?

I think the use of "value" is simply misleading in this particular case. We use "value" for quite a few things in the standard, including the "value" of template parameters (such a value might be a type). In this case, we're just looking at the expression and should avoid any idea that we're prematurely runtime-evaluating that expression in any way. I don't think a full-scope review is in order for these things; the entire [expr] section could do with a brush-up, but that's probably too large to cover in an editorial issue.

I just meant another opinion by someone from CWG, not necessarily by the actual group. Let me ask around.

Fixed.