@ahennequ has found out that the scaling by 1.e-3 of the dz variable in both the X fit and simultaneous fit was not ported properly to the cubic term. Fixing this allows to remove the loop in the X projection fit, reducing the number of calls by 2%.