Of the 22 columns, one is an identifier, six are categorical dimensions, thirteen are numeric metrics, and the only true date/time column is a technical ingestion timestamp. The classification below comes from the column profile and the distinct and range checks I ran (steps 0 to 4). **Identifier (1)** - `user_id` is a VARCHAR primary key. All 8,500 rows have a distinct value, running from USR-00001 to USR-08500 with no nulls. The profiler's "~7,297 distinct" is only an approximation, and the exact check gave 8,500 distinct values. **Categorical dimensions (6)** - `gender` (3 values), `occupation_type` (5), `chronotype` (3), `primary_bedtime_app` (6) and `sleep_debt_category` (4) are text classifiers. - `blue_light_filter_active` is stored as BIGINT but is a 0/1 flag with 2 values, so treat it as a dimension, not a metric. - `sleep_debt_category` is derived from `total_sleep_hours`. Keep it out of any model that predicts sleep outcomes from sleep hours. **Numeric metrics (13)** - Behaviour and exposure: - `bedtime_phone_minutes` (1–180) - `screen_brightness_pct` (10–100) - `caffeine_post_5pm_mg` (0–250) - `physical_activity_min` (0–112) - `age` (18–65) - Sleep outcomes: - `sleep_latency_min` (6.0–123.3) - `total_sleep_hours` (3.2–9.8) - `deep_sleep_pct` (8.1–28.0) - `rem_sleep_pct` (9.6–27.0) - Next-day outcomes: - `morning_alarm_snoozes` (0–7) - `next_day_fatigue_score` (1.0–10.0) - The profiler tags some of these as "categorical" (for example `morning_alarm_snoozes`, `next_day_fatigue_score`, `age`) only because they have few distinct values. They are still numeric measures. - `age` is a measure by role, though you could bin it into groups for segmentation. **Dates and times (1 true, 1 look-alike)** - `_ingestion_timestamp` is the only TIMESTAMP column. It holds a single value, 2026-10-01 03:08:19, so it carries no analytical time dimension. - `bedtime_phone_minutes` is labelled "temporal" by the profiler, but it is a duration in minutes, not a date or time. Treat it as a numeric metric. - The dataset has no event date, so there is no real time series to trend. **Technical lineage columns (3)** - `_batch_id`, `_source_file` and `_source_system` each have exactly 1 distinct value. They are ingestion metadata and carry no analytical value.