Hello!! It's Joseph

 Without any change to the polar scope, the error can be reduced to 8' to 10' ( 1/4 to 1/5 of 40 min polaris-NCP angular distance) . Some can achieve even smaller errors, YMMV. I am using TV aerial method set the polar scope, so that one of grub screw is at 12 o'clock. only adjust the other 2 grub screw in each cycle horizontal ( 3 o'clock , 6 o'clock) and then vertical direction( 12 o'clock and 6 o'clock). turn the scope 120 degrees, so that the other grub screw is at 12 o'clock repeat above steps as many as you can Now change the grub screw with M3 x 10mm thumb screw, the error is reduced  4' to 5 ' ( 1/8 to 1/10 of 40  min polar-NCP angular distance but still, when  you rotate the scope, the field is rotated together with the rotation. So far the main drawbacks of existing design are: Need to rotate the polar scope, which is held in place by 3 magnets. The polar scope arm flexure . There is no way to keep the polar scope

spec

shot at PTC 18~135mm zoom @132mm F5.6 ISO800 3 min x 7 subs resized to 25% to make the star without trails is always challenging done: polar scope is better collimated , so that error is 5' to 10' ( 1/4 to 1/5 of 40 min polaris-NCP angular distance) . Now using , done based on astronomyshed's method use sturdier tripod things to do: to further reduce the polar scope error, the scope still have "cone error" , reasons can be - the polar scope optics not in good optical alignment, - the polar scope arm is not always orthogonal to the polar scope, when the polar scope is rotated - something else. All in all, whether the polar scope arm is in parallel to RA axis of the mount, is another question  change the grub screw to big heard thumb screw use drift method to increase accuracy