The dataset has one identifier, six categorical dimensions, eleven numeric metrics and one real timestamp. The catalog tags its columns with a semantic role, and I adjusted a few of those tags where they misread the data. **Identifier** - `user_id` is a text ID running USR-00001 to USR-08500, with no nulls. The grain is one row per user. - The catalog's distinct count is only approximate (~7,297 against 8,500 rows), so I haven't confirmed that every ID is unique. **Categorical dimensions** - `gender` has about 3 values. - `occupation_type` has about 5 values, such as Corporate 9-to-5 and Student. - `chronotype` has about 3 values, such as Intermediate and Night Owl. - `primary_bedtime_app` has about 6 values, such as Instagram / Reddit and YouTube. - `sleep_debt_category` has about 4 values, from Mild Deficit to Severe Sleep Debt. It is derived from `total_sleep_hours`, so it is an outcome label rather than an independent input. - `blue_light_filter_active` is a 0/1 flag stored as BIGINT. Treat it as a binary category and don't average it as a measure. **Numeric metrics** - Demographic and behavioral inputs: - `age` (18–65) - `bedtime_phone_minutes` (1–180) - `screen_brightness_pct` (10–100) - `caffeine_post_5pm_mg` (0–250) - `physical_activity_min` (0–112) - Sleep and next-day 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) - `morning_alarm_snoozes` (0–7) - `next_day_fatigue_score` (1.0–10.0) The catalog tags `bedtime_phone_minutes` as temporal, but it is a duration in minutes, not a date or time, so it belongs with the metrics. It is also flagged as person-identifying, so aggregate it and don't quote individual values. **Dates / times** - `_ingestion_timestamp` is the only true timestamp. It holds a single value (2026-10-01 03:08:19), so it records when the data was loaded and cannot support trend analysis. - The dataset has no event dates, so there is no time dimension for the sleep behaviour itself. **Pipeline metadata (not analytical)** - `_batch_id`, `_source_file` and `_source_system` each hold one constant value. They are lineage fields and carry no analytical signal.