Question
You asked an AI assistant for a SQL query that computes monthly recurring revenue (MRR) from a subscriptions table, and it produced a clean, well-formatted query with CTEs that returns numbers in a believable range. You're not a billing-domain expert and the proration logic looks reasonable. Finance will use this for a board deck. How do you verify the query is actually correct before anyone trusts the output?
Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it — tests, edge cases, reading critically — and how you’d re-prompt or decompose to get it right.
Vibe coding: describe the solution in plain language (or narrate it) and the coach grades your approach. Generating runnable code from your description is coming next.