A couple of editorial cleanups would seem nice post-CWG2672:

1. In the writeup of CWG2672, I suggested: "[...] move the rule somewhere more general", which has not happened. The general rule that substitution into a lambda-expression considers the body to not be in the immediate context, which applies to all substitutions, is still in [temp.deduct.general], the section on template argument deduction.

2. The example in [temp.deduct.general]/9, immediately after the normative wording changed by that issue, doesn't match the rule. Prior to CWG2672, the rule was that all substitution into a lambda-expression during template argument deduction was not in the immediate context, and afterwards, the rule is that only substitution into the body is in the immediate context (but it doesn't matter if we're doing template argument deduction or something else). In particular, the h example is now valid:

template <class T>
  auto h(T) -> decltype([x = T::invalid]() { });
void h(...);
h(0);               // error: invalid expression not part of the immediate context

... should presumably say something like OK; invalid expression is in the immediate context.