Maybe this was discussed, but was a confidence intervals approach considered in the optimization? For instance, if the question were that I only need 40% confidence in my maximization objective, then strategies other than kelly (or in conjunction with kelly) could be used.