Solve 3 timed pandas cases: the screen rewards working code + narration, not perfect syntax.
Case 1 — messy orders (20 min)
Load the ecommerce orders CSV; fix dtypes, dedupe, build monthly revenue + top-10 products. Narrate each step: what you assume, what you check.
Case 2 — RFM mini (20 min)
Recency/frequency/monetary quintiles on the transactions file; name 2 segments and one action each. Pattern follows the RFM35 EDA playbook.
Case 3 — join trap (15 min)
Two files with mismatched keys (case + whitespace). Clean keys, left-join, report the unmatched rate. The trap is silent row loss — always count before and after the join.