First, I really appreciate your politeness in the discussion. One huge mis-use I can think of in using a Kelly type system would be to use past frequencies dir

First, I really appreciate your politeness in the discussion. One huge mis-use I can think of in using a Kelly type system would be to use past frequencies directly as your probabilities. At best this would give you the CAPM and at worst it would give you a disastrous "momentum trade" strategy. The Kelly System is a little sneaky in that it assumes you have good estimates of the probabilities of future events. So when the system fails it can turn around and say that you had bad estimates or un-modeled dependencies.