{
    "action": "tool_call",
    "reasoning": "No evidence from tool calls is visible this turn, so I'll re-verify the data-quality claims with one profiling query before answering: nulls, ID uniqueness, ranges, stress zeros, and the High School age inconsistency.",
    "tool": "query_sql",
    "args": {
        "sql": "SELECT COUNT(*) AS n_rows, COUNT(DISTINCT Student_ID) AS distinct_ids, COUNT(Academic_Performance_GPA) AS gpa_nonnull, COUNT(Perceived_Stress_Score) AS stress_nonnull, SUM(CASE WHEN Perceived_Stress_Score = 0 THEN 1 ELSE 0 END) AS stress_zero, SUM(CASE WHEN Academic_Level = 'High School' THEN 1 ELSE 0 END) AS hs_total, SUM(CASE WHEN Academic_Level = 'High School' AND Age >= 19 THEN 1 ELSE 0 END) AS hs_age19plus, SUM(CASE WHEN Daily_Usage_Hours + Sleep_Duration_Hours > 20 THEN 1 ELSE 0 END) AS time_over_20, MIN(Daily_Usage_Hours) AS min_use, MAX(Daily_Usage_Hours) AS max_use, MIN(Sleep_Duration_Hours) AS min_sleep, MAX(Sleep_Duration_Hours) AS max_sleep, MIN(Academic_Performance_GPA) AS min_gpa, MAX(Academic_Performance_GPA) AS max_gpa, MIN(Perceived_Stress_Score) AS min_stress, MAX(Perceived_Stress_Score) AS max_stress FROM raw.Social_media_impact_on_life",
        "purpose": "Verify data-quality findings: nulls, ID uniqueness, ranges, stress floor values, age vs. level inconsistency",
        "source": "raw"
    }
}