Thanks to all who responded. All are right. Sometimes only recalculating helps, removing the attachment of the sketch helps, etc. I've done some comparison between FC versions and the different solvers. I can confirm chrisb's analysis. When I use DogLeg the sketch is valid and makes a valid pad, when I use LM the curve is not closing well and yields an invalid sketch for padding. I've confirmed this by switching several times between the two solver algorithms.
Thanks again for your input!