{
    "action": "tool_call",
    "reasoning": "Check data quality: duplicate IDs, impossible or outlier numeric values after stripping non-numeric characters from the VARCHAR columns, and the ranges of the Productivity, Accuracy and Satisfaction scores.",
    "tool": "query_sql",
    "args": {
        "sql": "WITH t AS (SELECT User_ID, TRY_CAST(NULLIF(regexp_replace(Age,'[^0-9.]','','g'),'') AS DOUBLE) AS age, TRY_CAST(NULLIF(regexp_replace(Monthly_Income,'[^0-9.]','','g'),'') AS DOUBLE) AS inc, TRY_CAST(NULLIF(regexp_replace(AI_Usage_Hours_Per_Day,'[^0-9.]','','g'),'') AS DOUBLE) AS usage, TRY_CAST(NULLIF(regexp_replace(Work_or_Study_Hours_Per_Day,'[^0-9.]','','g'),'') AS DOUBLE) AS work, TRY_CAST(NULLIF(regexp_replace(Time_Saved_Hours_Per_Week,'[^0-9.]','','g'),'') AS DOUBLE) AS saved, TRY_CAST(NULLIF(regexp_replace(Monthly_AI_Cost,'[^0-9.]','','g'),'') AS DOUBLE) AS cost, Productivity_Score, Accuracy_Rating, Satisfaction_Score FROM raw.AI_Usage_and_Impact_on_Students_and_Professionals) SELECT COUNT(*) AS n, COUNT(DISTINCT User_ID) AS distinct_ids, COUNT(age) AS age_n, MIN(age) AS age_min, MAX(age) AS age_max, COUNT(*) FILTER (WHERE age<14 OR age>80) AS age_odd, COUNT(inc) AS inc_n, MAX(inc) AS inc_max, MEDIAN(inc) AS inc_med, COUNT(*) FILTER (WHERE inc=0) AS inc_zero, MAX(usage) AS usage_max, COUNT(*) FILTER (WHERE usage>work) AS usage_gt_work, MAX(work) AS work_max, COUNT(*) FILTER (WHERE work>16) AS work_gt16, MAX(saved) AS saved_max, COUNT(*) FILTER (WHERE saved>usage*7) AS saved_gt_weekly_usage, MAX(cost) AS cost_max, 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, COUNT(Productivity_Score) AS prod_n, COUNT(Accuracy_Rating) AS acc_n, COUNT(Satisfaction_Score) AS sat_n FROM t",
        "purpose": "Data-quality profile: duplicates, impossible values, outliers and score ranges",
        "source": "raw"
    }
}