It's a rounding thing internally it uses a 0.51ldu resolution to approximate the curve. So unless the donor part exactly lines up with a step it will be slightly off the analog curve.
Usually it isn't really that of a problem but the combination of the relative small link parts, a strong curve (radius<40) and the fact links have overlap makes it visible.
I'll see if things improve if increase the internal resolution.
edit, forgot about your other question.
You can rotate the links on the curve by rotating the first (circle) path point.
edit2: i just checked t's 1ldu stepping.
Usually it isn't really that of a problem but the combination of the relative small link parts, a strong curve (radius<40) and the fact links have overlap makes it visible.
I'll see if things improve if increase the internal resolution.
edit, forgot about your other question.
You can rotate the links on the curve by rotating the first (circle) path point.
edit2: i just checked t's 1ldu stepping.