At some point in the development cycle, you'll need to think about the criteria for choosing variables and values. These choices are so specific to the problem under consideration, sometimes even specific to the end-user, that there are no "canned" rules ready to use "off-the-shelf." You really must exploit professional expertise from the problem domain.

Solver treats optimization problems by enforcing the idea that the current solution must cost less than the preceding solution at each iteration.

The essential factor is in the propagation of the cost variable; at every moment, its boundaries must be the best evaluation possible of the cost of the solution partially examined.

We cannot say often enough that "optimality at all costs" is frequently unrealistic or even impossible. To get a good solution in a reasonable amount of time depends above all on the generation strategy. The behavior that you want, of course, is to get significantly better solutions very fast. With such an aim, it may be quite relevant to impose a rule such as, "the current solution must be 10% better than the next." Or, similarly, for example, it may be a good idea to eliminate once and for all the solutions close to the one already obtained by removing the values near the value belonging to the most recent solution.