What I find interesting is the seemingly uncritical acceptance of the NHTSA test result. What caused the drop in ratings (granted, it was on a test that is less demanding on vehicle structure than IIHS small overlap)? Maybe it was a flaw in the government's testing. Maybe the blame is Mazda's.
So you're saying Mazda was sending two different versions of 2014 CX-5 to NHTSA and IIHS for crash testing? Is it even illegally possible? If car manufactures can do this, then they can build-to-suit to get better crash ratings for different crash criteria?! Do you have a IIHS test date for 2014 CX-5?
Then again, Mazda added some reinforced structure at front driver side in its 2014 CX-5 only for IIHS's new driver side small overlap frontal crash test. OP's IIHS video also indicated this and that's why IIHS started new passenger side small overlap frontal crash test. How was that patched work on driver side affected that much on passenger side crash rating for CX-5 on NHTSA's full frontal crash test? And why other vehicles made the same patch work for better IIHS driver side small overlap frontal crash test rating didn't have ANY fallbacks on passenger rating in NHTSA's crash test? To me, there must be something else made by Mazda we simply just don't know.
If you think that the NHTSA star system information I quoted is out of date or for older pre-2011 HNTSA star system, the article is dated on 3/12/2016. Even if we use your definition for NHTSA's star system, I'd still say from "injury risk for this vehicle is much less than average" of a 5-star rating to "injury risk for this vehicle is average to greater than average" of a 3-star rating is a big drop off, not something we can say "negligible"!
No matter how you explained it, the fact remains. 2016 CX-5 does have 3-star passenger crash rating from NHTSA and 3-star is the worst frontal crash rating among all compact CUVs crash tested by NHTSA!