{
    "action": "final",
    "content": "## Column classification for `raw.Social_media_impact_on_life` (4,500 rows, 20 columns)\n\nThis is based on the column metadata and the profiling queries I ran (steps 0–4).\n\n### Identifier\n- **Student_ID** (VARCHAR): primary key. 4,500 rows and 4,500 distinct IDs, no nulls, so it is one row per student. The catalog's \"~4717 distinct\" is only an approximation; the exact count is 4,500.\n\n### Categorical dimensions (group-by / segment fields)\n- **Age** (BIGINT, 15–26, 12 distinct values). It is numeric but behaves as a categorical age band. The catalog tags it as a \"measure\" in one view and a \"dimension\" in another. I would treat it as a dimension, since there are only 12 distinct values and the gold tables group by it.\n- **Gender** (4 values, including \"Prefer not to say\")\n- **Academic_Level** (3 values, High School to Undergraduate)\n- **Primary_Platform** (7 values, Instagram to YouTube)\n- **Device_Type** (3 values, Laptop/PC to Tablet)\n- **Social_Comparison_Frequency** (5 values, Always to Sometimes)\n- **Overall_Impact** (3 values, Beneficial to Neutral). This is an outcome label, so it is better treated as a target or outcome category than as a predictor.\n- **Late_Night_Usage** (BOOLEAN, true/false). This is a binary flag.\n- **Sleep_Quality_Score** (BIGINT, 1–5, 5 distinct values). It is a rating scale, so it can be read as ordinal categorical or as a numeric metric. I list it under metrics because it is averaged, but it is only a 5-point scale.\n\n### Numeric metrics (continuous or quantitative measures)\n- **Daily_Usage_Hours** (DOUBLE, 0.9–14.0). The catalog labels it \"temporal\", but it is a duration in hours, not a date, so it is a numeric metric.\n- **Weekend_Extra_Hours** (DOUBLE, 0.0–4.5). Same point: a duration, not a date.\n- **Sleep_Duration_Hours** (DOUBLE, 3.0–10.5)\n- **Sleep_Quality_Score** (BIGINT, 1–5). See the note above.\n- **Perceived_Stress_Score** (DOUBLE, 0–40). 46 nulls.\n- **Mental_Health_Index** (BIGINT, 32–98). No nulls.\n- **Academic_Performance_GPA** (DOUBLE, 1.9–4.0). 85 nulls.\n\n### Dates / times\n- The business data has **no true date or time column**. The \"TEMPORAL AXES\" on the card (Daily_Usage_Hours and Weekend_Extra_Hours) are hour durations, so there is no real time series. Trend-over-time questions cannot be answered from this dataset.\n- The only real timestamp is **_ingestion_timestamp** (TIMESTAMP). It has a single distinct value, 2026-10-01 14:08:19, so it records when the data was loaded and carries no analytical signal.\n\n### Pipeline metadata (not analytical)\n- **_batch_id**, **_source_file** and **_source_system** each have a single distinct value. They are load-lineage columns and can be ignored for analysis.\n\n### Data-quality notes\n- Only two metrics have missing values: Perceived_Stress_Score (46 nulls) and Academic_Performance_GPA (85 nulls). Every other column is fully populated.\n- Student_ID is unique, so there are no duplicate students."
}