<invoke name="tool_call"> <parameter name="reasoning">Check how the key categorical columns split the records, starting with User_Type crossed with Education_Level (card says ~17 distinct education values, which may be dirty).</parameter> <parameter name="tool">query_sql</parameter> <parameter name="args">{"sql": "SELECT User_Type, Education_Level, COUNT(*) AS n, ROUND(AVG(Productivity_Score),2) AS avg_productivity, ROUND(AVG(Satisfaction_Score),2) AS avg_satisfaction FROM raw.AI_Usage_and_Impact_on_Students_and_Professionals GROUP BY User_Type, Education_Level ORDER BY n DESC", "purpose": "Profile User_Type x Education_Level segments with size and outcome averages", "source": "raw"}</parameter> </invoke>