Coming from NIH-land and occasionally dabbling in NSF-funding via EEID, RAPID and some collaborations with CS folks, the broader impacts section remains the thing that's most opaque to me.