A complete regression check needs a baseline, a learned model and an error measure that matches the decision. Work through fresh fictional dispatch records in this practice set. Every task supplies its own training and test objects, so treat each as a separate analysis rather than carrying estimates from an earlier answer.
The baseline tasks require the training mean. The learned model tasks require an intercept, and the metric tasks require the original duration units. Squared error and root mean squared error are easy to confuse because both start by squaring residuals. MAE uses absolute values instead. Read the metric name before selecting a function. You don't need another estimator to finish these tasks; the work is making the existing workflow consistent and reproducible.