I joined five tables of an e-commerce dataset with SQL to rank suppliers by revenue, then stress-tested the number before trusting it. Found that 69% of “total revenue” came from a block of rows with an artificial, non-random pattern, and the supplier ranking fell apart once corrected for it.