Question
An AI agent refactored a 600-line Python billing module (proration, tax rounding, currency conversion) into smaller functions. Your existing unit suite still passes green, and a diff review looks like a faithful structural rewrite. But you didn't write the original line-by-line and proration math is subtle. Before merging, how do you actually verify behavior is unchanged — not just that the tests you happen to have still pass?
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.