The delegating nodes are deprecated since 2.10. Please use the Recursive Loop Nodes instead. The idea of delegating is to build a new model only for data points being incorrectly classified/predicted by the model. Hence, a collection of models is achieved. The loop start therefore is initialized with the whole data. The loop end node receives only wrongly classified data in each iteration. This will be passed back to the loop start, which outputs the data again, until no more data points are inserted to the loop end.