The table `raw.bedtime_screentime_sleep_debt` has 22 columns: the 18 in the profiling result plus 4 ingestion-metadata columns from the dataset card. Types and ranges come from the tool output. The meanings are my reading of the column names, since the dataset has no written data dictionary. **Identifier and demographics** - `user_id` (VARCHAR): unique ID, one row per person. - `age` (BIGINT): age in years, from 18 to 65. - `gender` (VARCHAR): categorical, about 3 values. - `occupation_type` (VARCHAR): categorical, about 5 values. - `chronotype` (VARCHAR): whether the person is a morning or evening type, about 3 values. **Bedtime phone and screen behavior** - `bedtime_phone_minutes` (BIGINT): minutes of phone use at bedtime, from 1 to 180. The card flags it as person-identifying, so it should only be aggregated, never quoted per person. - `primary_bedtime_app` (VARCHAR): the main app used at bedtime, about 6 values. - `screen_brightness_pct` (BIGINT): screen brightness as a percentage, from 10 to 100. - `blue_light_filter_active` (BIGINT): a 0/1 flag for whether a blue-light filter was on. **Evening lifestyle** - `caffeine_post_5pm_mg` (BIGINT): caffeine consumed after 5pm in mg, from 0 to 250. - `physical_activity_min` (BIGINT): minutes of physical activity, from 0 to 112. **Sleep outcomes** - `sleep_latency_min` (DOUBLE): minutes taken to fall asleep, from 6 to 123.3. - `total_sleep_hours` (DOUBLE): total sleep duration in hours, from 3.2 to 9.8. - `deep_sleep_pct` (DOUBLE): share of sleep spent in deep sleep, from 8.1 to 28. - `rem_sleep_pct` (DOUBLE): share of sleep spent in REM, from 9.6 to 27. - `sleep_debt_category` (VARCHAR): a label with about 4 values. The card says it is computed from `total_sleep_hours`, so reuse it rather than re-deriving it. **Next-day effects** - `morning_alarm_snoozes` (BIGINT): number of alarm snoozes, from 0 to 7. - `next_day_fatigue_score` (DOUBLE): fatigue rating from 1 to 10. **Pipeline metadata** (listed on the card, not in the profiling result) - `_ingestion_timestamp` (TIMESTAMP): when the row was loaded. - `_batch_id`, `_source_file`, `_source_system` (all VARCHAR): load lineage. No column has missing values, since every null count in the profile is 0. The profiler's automatic type labels are sometimes off. For example, it tags `age` as a "duration" and `bedtime_phone_minutes` as "temporal", though neither is a date. Treat the physical types above as authoritative.