Tree to Rules and Classification Error

The tree inside "Tree to Rules" is the copy-and-paste of "Tree W".

Both the operators, Tree W and Tree to Rules, are passed to the same Apply Model and Performance operators (just the names are different, they are copy-and-paste, so the same configurations).

How is it possible that the Classification Error of Tree W is 5.59% and that of Tree to Rules is 0.79%?

They have the same input, they go through the same performance test and Tree W is also executed inside Tree to Rules.

The only difference, visually speaking, is that Tree to Rules is kind of overfitting generating tons of rules while Tree W don't.

How is it possible?