Your question is Time-Series ML Workflow Changes. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
They challenged your classification approach and asked whether the problem should instead be treated as a forecasting problem; how would the ML workflow change, including train/test splitting, time dependencies, and evaluation for time-series data?