Your feedback loops should be short enough that the cost of making a mistake is low enough that the value of the learning in the feedback loop exceeds the magnitude of the cost of a mistake. In other words: shorten your feedback loops to the point that you're no longer afraid to make a mistake.