{
    "action": "tool_call",
    "reasoning": "No query results are in evidence yet, so the figures in my earlier draft are unverified. I'll profile data quality with one SQL query: row count, distinct IDs, blank and null counts per column, and the numeric columns' ranges.",
    "tool": "query_sql",
    "args": {
        "source": "raw",
        "purpose": "Profile data quality: duplicates, blanks, and ranges of the numeric measures",
        "sql": "SELECT COUNT(*) AS n_rows, COUNT(DISTINCT User_ID) AS distinct_users, COUNT(*) - COUNT(NULLIF(TRIM(Age),'')) AS age_blank, COUNT(*) - COUNT(NULLIF(TRIM(Gender),'')) AS gender_blank, COUNT(*) - COUNT(NULLIF(TRIM(Country),'')) AS country_blank, COUNT(*) - COUNT(NULLIF(TRIM(Monthly_Income),'')) AS income_blank, COUNT(*) - COUNT(NULLIF(TRIM(Monthly_AI_Cost),'')) AS cost_blank, COUNT(*) - COUNT(NULLIF(TRIM(AI_Usage_Hours_Per_Day),'')) AS usage_blank, COUNT(*) - COUNT(NULLIF(TRIM(Time_Saved_Hours_Per_Week),'')) AS timesaved_blank, COUNT(*) - COUNT(NULLIF(TRIM(Work_or_Study_Hours_Per_Day),'')) AS workhrs_blank, COUNT(*) - COUNT(Productivity_Score) AS prod_null, COUNT(*) - COUNT(Accuracy_Rating) AS acc_null, COUNT(*) - COUNT(Satisfaction_Score) AS sat_null, MIN(Productivity_Score) AS prod_min, MAX(Productivity_Score) AS prod_max, MIN(Accuracy_Rating) AS acc_min, MAX(Accuracy_Rating) AS acc_max, MIN(Satisfaction_Score) AS sat_min, MAX(Satisfaction_Score) AS sat_max, MIN(TRY_CAST(REGEXP_REPLACE(Age,'[^0-9.]','','g') AS DOUBLE)) AS age_min, MAX(TRY_CAST(REGEXP_REPLACE(Age,'[^0-9.]','','g') AS DOUBLE)) AS age_max, MAX(TRY_CAST(REGEXP_REPLACE(AI_Usage_Hours_Per_Day,'[^0-9.]','','g') AS DOUBLE)) AS usage_max, MAX(TRY_CAST(REGEXP_REPLACE(Work_or_Study_Hours_Per_Day,'[^0-9.]','','g') AS DOUBLE)) AS workhrs_max FROM raw.AI_Usage_and_Impact_on_Students_and_Professionals"
    }
}