Woooooo, I finally figured out what is causing the twitch.
Regrettably, it looks the rotation lerp converts the normalized control rotation to a standard rotation internally before the lerp takes place. It turns out that my earlier assumption that Rlerp would preserve the normalized control rotations at all axes at +/- 180 degree was incorrect. This is why, from certain angles in the 360 degree pawn rotation, the mashed rotation for the spring arm won’t lerp smoothly between states 0 and 1.
It’s not quite a bug, because my use case (mashing 3 Rinterps together to get spring arm lag that can vary on different axes… and then lerping the result) is unusual.
By themselves, the mashed 3 Rinterps work fine, because they preserve the format. Yes, I’m obsessed with figuring this out 
My investigations into a solution continue!