If your tool returns rows, your model just became the database: you pushed…

If your tool returns rows, your model just became the database: you pushed counting, filtering and aggregation into the LLM and made correctness depend on how many tokens it will spend. The Kaggle runs show a tool that returns counts produces correct answers at a flat, small token cost; a rows tool splits models into those that painstakingly count (and burn tokens) and those that answer confidently but wrong. Return computed answers or typed summaries from tools, not raw result sets. Put the count, min and max in the tool; leave arithmetic off the model unless you accept the token bill and instability.