A query can run perfectly and answer the wrong question.

Our tiny fictional CSV returned 900 for paid October orders. Remove one status filter and the total becomes 1,800. Same file. Different question.

Try the starter:
1. Write the question before opening a chatbot.
2. List your column names and types.
3. Add the date window and which statuses count.

Then pause. The missing pieces are the reviewed SQL, explicit CSV types, and a result you can check by hand.

The full AI Anonymous walkthrough includes a local DuckDB exercise, both authored queries, the actual observed results, and a proposed schema-only AI prompt:
https://jabsystems.io/ai-anonymous/duckdb-check-the-math

We tested the SQL exercise on fictional data. We did not test AI-generated SQL, model quality, or speed. Review generated SQL as code before running it; local execution is not a sandbox.

Save this for the next spreadsheet question that needs a checkable answer.

AI Anonymous · from JAB Systems
#AIAnonymous #DuckDB #PracticalAI #DataLiteracy #SQL