This thread is about GFCI, but the AFCI mess has all kinds of studies that conflict one another's results. I'm not totally picking a side, but the fact that there is legitimate studies that oppose one another tells me they may at least have jumped the gun on implementing code before they had it figured out, and still may not have it figured out.
you will always have
scenarios where it works
and where it doeesn't
do you forsake those it may save because sometimes it doesn't work
like airbags or seat belts
it's probability based