{
    "action": "tool_call",
    "reasoning": "The question appears to be about classifying the table's columns. My earlier draft cited catalog approximations (distinct Student_ID, null counts, ranges) that I never verified with a query. I'll run one profiling query to confirm them before answering.",
    "tool": "query_sql",
    "args": {
        "sql": "SELECT COUNT(*) AS n_rows, COUNT(DISTINCT Student_ID) AS distinct_ids, COUNT(Student_ID) AS nonnull_ids, MIN(Age) AS min_age, MAX(Age) AS max_age, COUNT(DISTINCT Age) AS distinct_ages, COUNT(DISTINCT Gender) AS distinct_gender, COUNT(DISTINCT Academic_Level) AS distinct_level, COUNT(DISTINCT Primary_Platform) AS distinct_platform, COUNT(DISTINCT Device_Type) AS distinct_device, COUNT(DISTINCT Social_Comparison_Frequency) AS distinct_comparison, COUNT(DISTINCT Overall_Impact) AS distinct_impact, COUNT(*) - COUNT(Perceived_Stress_Score) AS stress_nulls, COUNT(*) - COUNT(Academic_Performance_GPA) AS gpa_nulls, COUNT(*) - COUNT(Mental_Health_Index) AS mhi_nulls, MIN(Perceived_Stress_Score) AS min_stress, MAX(Perceived_Stress_Score) AS max_stress, MIN(Mental_Health_Index) AS min_mhi, MAX(Mental_Health_Index) AS max_mhi, MIN(Academic_Performance_GPA) AS min_gpa, MAX(Academic_Performance_GPA) AS max_gpa, MIN(Sleep_Duration_Hours) AS min_sleep, MAX(Sleep_Duration_Hours) AS max_sleep, MIN(Sleep_Quality_Score) AS min_sq, MAX(Sleep_Quality_Score) AS max_sq, COUNT(DISTINCT _ingestion_timestamp) AS distinct_ingest_ts FROM raw.Social_media_impact_on_life",
        "purpose": "Verify key uniqueness, dimension cardinalities, null counts, and metric ranges to support the column classification",
        "source": "raw"
    }
}