You don't seem to know at all what you are talking about. Divergence is not always computed in only the way you seem to suggest. Divergence behavior of bounded oscillator-type indicators does appear on many different indicators. Is it useful? It can be very powerful. Does it fail? Of course it does. But perfection is not something attainable anyway in any indicator. It is a matter of proportion of its successes. How you decide to classify the occurrence of divergence requires many decisions..how long is the lookback period, are equal indicator extremes gong to be included, how many bars apart are accceptable, what about situations where a third less extreme point follows the two..this usually implies that the first divergence trade will be somewhat underwater, but this next extreme will often work. The less lag, the greatest smoothness and the retention of sharpness at the turn in the indicator assists in the divergence signal being timely and usefull. Classifying divergence in real-time manually by the trader is difficult in the heat of battle. The computer algorithms hae improved over the years to perform this classification. This presents an objective rule-base way to do this but it still requires all the preceeding decisions to be made and coded or entered as inputs by the user. Important tops and bottoms often are accompanied by clear divergence; it is a matter of filtering out the erroneous signals by some means...which is pretty much the problem we face with every trading approach.