"Currently there are some limitations for shift pattern rules: The all-or-nothing match semantic as a hard constraint cannot be achieved on a larger scale, especially if we can expect rules defined for (almost) every employee in the dataset. The solver cannot incrementally improve the match, as unassigning any shift that breaks the pattern always leads to a better score. In fact, it would be better that this feature is changed, so that it is not a hard constraint (it's not a limitation of the real world or a legislation that always has to be fully respected), it's rather a strong soft constraint."