The process in rule-based systems that constructs a hypothesis, then tests it by working backward through the rules to see if the hypothesis is supported. An example in weather forecasting is to assume there will be afternoon thunderstorm activity, then determine whether the data support this assumption.