See #457. There were more sparse masters required than I expected initially, because in some cases the width of semibold layers is quite different from what one would expect as interpolation. I've preserved sparse masters for those where the width deviates by more than 10 units (as well as those where the outlines are visually different, obviously), but there will still be small width differences when the width of the semibold layer is considered close enough (<10 units) to the interpolated equivalent, but not exactly equal.
See #457. There were more sparse masters required than I expected initially, because in some cases the width of semibold layers is quite different from what one would expect as interpolation. I've preserved sparse masters for those where the width deviates by more than 10 units (as well as those where the outlines are visually different, obviously), but there will still be small width differences when the width of the semibold layer is considered close enough (<10 units) to the interpolated equivalent, but not exactly equal.