## Column classification (20 columns in `raw.Social_media_impact_on_life`) This is based on the column metadata I pulled. The profiler's labels are sometimes off, so I've corrected them where they don't fit the data. **Identifier (1)** - `Student_ID` (VARCHAR) is the primary key, with one row per student and no nulls. The range is `STU_2024000` to `STU_2028499`. The profiler's distinct count (~4,717) is only an approximation, since the table has 4,500 rows. **Categorical dimensions (8)** - `Gender` has about 4 values, including "Prefer not to say". - `Academic_Level` has about 3 values: High School, Undergraduate and one other. - `Primary_Platform` has about 7 values, from Instagram to YouTube. - `Device_Type` has about 3 values, including Laptop/PC and Tablet. - `Social_Comparison_Frequency` has about 5 values, from Always to Sometimes. It is ordinal. - `Overall_Impact` has about 3 values, including Beneficial and Neutral. It is an outcome label. - `Late_Night_Usage` is a BOOLEAN flag. - `Age` is stored as BIGINT (15–26) and the profiler tagged it a "measure". It has only about 10 distinct values, so I'd treat it as a grouping dimension. That is how the gold tables use it, and you can still average it if needed. **Numeric metrics (9)** - `Daily_Usage_Hours` (0.9–14.0) and `Weekend_Extra_Hours` (0.0–4.5) are durations in hours. The profiler tagged them "temporal", but they are quantities of usage, not points in time. - `Sleep_Duration_Hours` (3.0–10.5). - `Sleep_Quality_Score` (1–5) is an ordinal rating that behaves like a small-scale score. - `Perceived_Stress_Score` (0–40) has 46 nulls. - `Mental_Health_Index` (32–98) has no nulls. - `Academic_Performance_GPA` (1.9–4.0) has 85 nulls. **Dates/times (effectively none for analysis)** - The dataset has no real event-date column, such as a survey or observation date. A time-trend analysis isn't possible. - `_ingestion_timestamp` (TIMESTAMP) is the only date/time column. It holds a single constant value (2026-10-01 14:08:19), so it records when the data was loaded, not when anything happened. **Pipeline metadata (3, not analytical)** - `_batch_id`, `_source_file` and `_source_system` each hold one constant value. They are lineage fields only. **Practical takeaway:** use the 8 dimensions to segment, and the 9 numeric columns as the measures to compare. The two "temporal" hour columns are usage-intensity measures, and the only nulls to handle are in stress score (46) and GPA (85).