But what do you do if you don't win? Then the target level is two epsilon away -- do you bet that amount?
Maybe that works if the probability of success is high enough and the initial epsilon is small enough. If that is not the case, it starts to sound like a martingale type strategy.
If you start super close to the goal, then yes you would double for every loss... for a while. When it starts actually cutting into your bankroll the bets would grow slower and slower and then even start shrinking. It should not surprise me if, far from the goal, you would bet a fraction of your bankroll every time, like Kelly does.
Maybe that works if the probability of success is high enough and the initial epsilon is small enough. If that is not the case, it starts to sound like a martingale type strategy.