A model comparison is easier to interpret when the only intended difference is the estimator. These tasks provide fixed training and test objects, then ask you to fit a stated tree or ensemble. The fictional records change between tasks, so carry the workflow rather than copying earlier predictions.
Classification and regression trees share a splitting idea but produce different kinds of outputs. Forests average randomized trees; boosting builds an additive sequence. Keep the estimator family, size, depth and seed requested in each prompt. Those settings make the comparison repeatable and keep fitting responsive in the notebook. The deliberate mistakes include fitting on held-out rows and substituting a dummy estimator. Both can return arrays of the right shape while answering the wrong experiment.