Is there a reliable training method to minimize the need for "debugging" in data sets like this?
Like: Here is a list of all stitches (so it doesn't create its own), and the number of stitches must be correct for dimensions, or something like that. Then you validate all the input data against that, and move on from there.