Design Delivery ETA Prediction

Delivery ETA is the regression question: a continuous target with a skewed distribution, error costs that are brutally asymmetric, and a prediction that changes the very behavior it is trying to predict, because users decide whether to order and couriers decide how to move based on the number you display. It is a staple at DoorDash, Uber, and Instacart, and it is where interviewers find out whether a candidate knows any loss function beyond MSE. This walkthrough follows the Interview Framework for food-delivery ETA at a marketplace platform, with one announced deviation: this is a constraint-dominated problem, so serving constraints come forward into Phase 2 and shape the modeling discussion instead of waiting for Phase 5. Say that reordering out loud; a named deviation reads as judgment.