Question
You ask an AI to write a Python function to compute the volume-weighted average price (VWAP) of trades from a Pandas DataFrame. It returns `(df['price'] * df['volume']).sum() / df['volume'].sum()` — correct for one symbol. Your data has 40 symbols interleaved and an hour of trades, and you actually need a *rolling intraday* VWAP per symbol that resets each day. You re-prompt 'do it per symbol' and it adds a `groupby('symbol')` but it's still a single number per symbol, not the running series you need. How do you re-steer to the right shape?
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.