Profile before you model: shape, types, target balance, leakage check. Then the part most candidates botch, that a missing value is often the most predictive feature in the table, so you flag it before you fill it. The first-hour playbook and the imputation decisions that survive production.
Updated Sep 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.
Profile before you model: shape, types, target balance, leakage check. Then the part most candidates botch, that a missing value is often the most predictive feature in the table, so you flag it before you fill it. The first-hour playbook and the imputation decisions that survive production.
The screen is whether the candidate has a real first-hour EDA routine (profile, distributions, target, leakage) AND understands that missingness is frequently informative rather than noise to be filled away. The strong move is to add a missing-indicator column before imputing, and to ask WHY the data is missing (MCAR vs MAR vs MNAR) because that decides the strategy. The reserved follow-up: 'you median-imputed income and the model got worse, why': usually because missing income was itself the signal and you erased it.
No comments yet — be the first to share your approach.
