This thread is on the right track! As a quick late at night investigation, I took the single screw out from each rear arm near the pivot without taking anything else apart. I thought, if there is enough wire slack maybe I could pop the arm off, give a few rotations, and put back. What I found is there was barely enough wire slack to slide the arm off and see a tiny bit of wire. Not enough that I felt comfortable spinning the arm. (maybe could have though) I pushed each arm back on, tightened the screw, and went for a flight for no good reason. What I found was that no matter how hard I tried, I could not get compass error and could not get redundancy switch. Even going sport mode, full elevator + full throttle at same time, no compass error. Now in the past I would easily have gotten compass switch every time. I proceeded to fly two full batteries in sport mode, not able to get compass to switch.
Now then using
@BudWalker 's miraculous and life saving CsvView, I compared the rear compass noise under full speed sport mode to previous .dat captures I had. The plots are below, and you can see previously I had about 250 counts pk2pk on rear magZ.Back and I now have about 150 counts pk2pk.
This tells me the compass interference is highly sensitive to the routing of the small amount of wire slack inside the Mavic body. Most likely all I did was slightly move the wire by pulling on it. I'm not sure my "fix" will last, and you can see front compass is still much cleaner than rear. However I'd be very confident now that just adding some twists to the rear motor wires would totally fix the issue.
Old .dat prior to my wire pull (full speed sport mode)
View attachment 6710
New flight after my wire pull (full speed sport mode)
View attachment 6712