Someone sent me this link: http://www.cs.clemson.edu/~steve/CPLSCMODS/stealth.note. I find this difficult to believe. I make lots of errors, but when developing a model I run a model in all its variants so many times, that I really should have some idea about the optimal objective or at least a good objective value. It would be very difficult to imagine that I would be so clueless about the optimal objective value that I can not detect an error like this just by looking at the solution. E.g. a hint something is wrong would be the observation that the final solution is worse than the initial point. The only exception would be if the model is so tight that the minimizer and maximizer yield essentially the same objective value, in which case the problem is not that severe.