Would it be valuable to compare these upticks to the size of gap after a prime n given n+2 is known to be notprime?
I wonder how much is a sort of 'magical' clearing out of primes around twins, versus the simple explanation that removing a possible gap of 2 from the distribution of nextprimegap alters the averages/medians in this way.
