Filtering logic for

The short-term aspect of this thread is what to do in 1.2. The
longer-term aspect is how should we change our process.

Su-Laine and Jeff have spoken to the latter, that "we should require one
or two implementations on the ground before including a new feature in
the spec."

Do we want to mandate this for every feature, or only for those that
require complex changes in processing? For example,