Even without the collar the hardware calibration from the factory is all over the place. The sample that just died easily hit the full range to the point I could reach 0/0 and 255/255 in each corner on the raw inputs with a few degrees of deflection to spare. The one I'm using now maxes out around ~230 in each axis with the collar bored out to allow full throw. Judging from the tiny range on the newest sample I tested, it wouldn't even reach 200 without the collar.