**Classification of the 20 columns in `raw.Social_media_impact_on_life`** (from the column metadata and the profiling query on 4,500 rows) **Identifier (1)** - `Student_ID` (VARCHAR) is the primary key. The profile query found 4,500 rows, 4,500 distinct IDs and no nulls, so there is one row per student. The catalog's "~4717 distinct" is only an approximation. **Categorical dimensions (8)** - `Gender` has 4 distinct values. - `Academic_Level` has 3: High School, Undergraduate and one other. - `Primary_Platform` has 7. - `Device_Type` has 3. - `Social_Comparison_Frequency` has 5, from Always to Sometimes. - `Overall_Impact` has 3: Beneficial, Neutral and one other. - `Late_Night_Usage` is a BOOLEAN with 2 values. - `Age` (BIGINT, 15–26, 12 distinct values) is stored as a number, but the catalog tags it as a dimension. It works best as a grouping field (age bands), though it can also be treated as a numeric covariate. **Numeric metrics (8)** - Outcome measures: - `Mental_Health_Index` is an integer from 32 to 98 with no nulls. - `Perceived_Stress_Score` runs from 0 to 40 and has 46 nulls. - `Academic_Performance_GPA` runs from 1.9 to 4.0 and has 85 nulls. - `Sleep_Duration_Hours` runs from 3.0 to 10.5. - `Sleep_Quality_Score` is an ordinal score from 1 to 5, with only 5 distinct values. Treat it as a rating scale rather than a continuous measure. - Usage measures (duration quantities, not dates): - `Daily_Usage_Hours` runs from 0.9 to 14.0. - `Weekend_Extra_Hours` runs from 0.0 to 4.5. - The catalog tags both as "temporal", but they are hours of use, so treat them as numeric exposure metrics. They are the natural predictors for the sleep, stress, mental-health and GPA outcomes. **Dates and times (effectively none)** - The data has no business or event date, so trend-over-time questions can't be answered from it. - The only true timestamp is `_ingestion_timestamp` (TIMESTAMP). It has a single distinct value (2026-10-01 14:08:19), so it is load metadata and not an analytic time axis. **Pipeline metadata (3)** - `_batch_id`, `_source_file` and `_source_system` each have one constant value. They are lineage fields with no analytic value. **Data-quality notes** - Only `Perceived_Stress_Score` (46 nulls) and `Academic_Performance_GPA` (85 nulls) have missing values. Every other analytic column is complete. - The automatic role tags are partly unreliable: they label `Age` a "measure", the two usage-hour columns "temporal", and several numeric columns "categorical". Use the classification above instead.