Thanks for the article! QQ: does dependence between data splits matter? If we bootstrap, we're testing and training on the same data, right?