Data & SQL

The expensive mistakes in data work are rarely syntax errors. They are the join that silently doubled every row, the NULL that quietly dropped a fifth of the customers, and the number in a deck that nobody can reproduce six weeks later. These prompts are built around that: they ask for the failure mode before the answer, insist on a row-count sanity check, and require the model to state its assumptions about your schema rather than inventing columns and moving on. Several are deliberately run before you execute anything.

36 prompts in 5 groups