> That fails if you do a first-order implementation because you have no way of knowing whether the person voting/sharing had already read the article, but not necessarily shared it at the time of reading.

It might still hold for a large group of people, though, and be valid in a statistical sense? Which is what they're really after.