{
    "action": "tool_call",
    "reasoning": "I need to examine the data quality systematically. The evidence so far shows percentage values in VARCHAR columns (Overall, Gender, Age, Income, etc.). I should check: 1) whether these percentages parse correctly as numbers, 2) if they sum appropriately within question groups, 3) if there are any values outside 0-100% range, 4) consistency of formatting (all appear to have % suffix), and 5) any NULL or missing patterns. Let me start by profiling the percentage columns to detect outliers, range violations, and format issues.",
    "tool": "query_sql",
    "args": {
        "sql": "SELECT COUNT(*) as total_rows, COUNT(DISTINCT Question) as distinct_questions, COUNT(DISTINCT Response) as distinct_responses, MIN(CAST(REPLACE(Overall, '%', '') AS DECIMAL)) as min_overall_pct, MAX(CAST(REPLACE(Overall, '%', '') AS DECIMAL)) as max_overall_pct, AVG(CAST(REPLACE(Overall, '%', '') AS DECIMAL)) as avg_overall_pct, COUNT(CASE WHEN CAST(REPLACE(Overall, '%', '') AS DECIMAL) < 0 THEN 1 END) as negative_pct, COUNT(CASE WHEN CAST(REPLACE(Overall, '%', '') AS DECIMAL) > 100 THEN 1 END) as over_100_pct, COUNT(CASE WHEN Overall NOT LIKE '%\\%%' THEN 1 END) as missing_pct_symbol FROM raw.polling_on_ai_usage_jul_2026",
        "purpose": "Check basic data quality: row counts, percentage value ranges, and format consistency",
        "source": "raw"
    }
}