Yeah, having read it I can't help but feel that a simulated annealing approach might end up generating a better result. If only because the limiting factor for

Yeah, having read it I can't help but feel that a simulated annealing approach might end up generating a better result. If only because the limiting factor for simulated annealing is how quickly you can generate random duties and check some objective function, and not how quickly you can solve some NP complete problem for a large number of variables. Of course having found a couple of good combinations of duties with simulated annealing there's nothing stopping you from trying the integer program on those (+ variations) to try and find a better version.