the difference between the actual outcomes and the current predictions. 3. Fit a decision tree to these residuals. 4. Add this tree to the ensemble (scaled by a learning rate). 5. Repeat steps 2--4 for a specified number of iterations.