In the argument list for shgo (https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.shgo.html) describing constraints, it reads as follows:

g(x) <= 0 applied as g : R^n -> R^m

This would imply that g(x) should be non-positive in order to satisfy the constraint. However, the cattle-feed example in the documentation indicates that the functions for the inequality constraints should be non-negative, with g1(x) and g2(x) explicitly indicated as being >= 0.

To make things even more confusing, in the same cattle-feed example, the values of g1(x) and g2(x) at the minimizing value res.x are shown to be negative, implying that the constraints aren't satisfied.