Always hold out data for validation. Backtested performance on the same data used for training is meaningless. - Be skeptical of complex models. Simple models with a few strong features often outperform complex models, especially with limited training data. - Quantify model uncertainty. A model that