Kelly criteria is wonderful if you have a static game. Where the odds don’t change. The Kelly criteria can fail miserably when the odds change. So that’s why most sophisticated gamblers employ a proportional Kelly because there is uncertainty around the odds, otherwise you can overbet. You need to scale down the Kelly bet based on the uncertainty around the win odds. Some just use a heuristic called the ‘half kelly’ where they divide the Kelly bet by half. But if you have a way to compute the variance around the win rate (data+computer) you can get a little more sophisticated.