Of the 22 columns, 1 is an identifier, 7 are categorical dimensions, 11 are numeric metrics, and only one is a true timestamp. This comes from the column metadata I pulled (type, role, null count, distinct count, range). I did not run any SQL, so there is no table to render. **Identifier (1)** - `user_id` (VARCHAR) runs from USR-00001 to USR-08500 with no nulls. The card says it is the key, with one row per user. The metadata's ~7,297 distinct count is an approximation and sits below the 8,500 rows. I have not checked for duplicate IDs with an exact `COUNT(DISTINCT)`. **Categorical dimensions (7)** - `gender` has ~3 values. - `occupation_type` has ~5 values, e.g. "Corporate 9-to-5" and "Student". - `chronotype` has ~3 values, e.g. "Intermediate" and "Night Owl". - `primary_bedtime_app` has ~6 values, e.g. "Instagram / Reddit" and "YouTube". - `sleep_debt_category` has ~4 values, from "Mild Deficit" to "Severe Sleep Debt". The card says it is derived from `total_sleep_hours`, so don't use the two together as independent predictors. - `blue_light_filter_active` is a 0/1 flag stored as BIGINT, so treat it as a binary category rather than a quantity. **Numeric metrics (11)** - `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_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) None of these has nulls. **Dates and times (1)** - `_ingestion_timestamp` (TIMESTAMP) holds a single value, 2026-10-01 03:08:19, for every row. It records when the data was loaded, not when any sleep event happened. **Things to watch** - **`bedtime_phone_minutes`:** the profiler tagged it as the "temporal" axis, and the card lists it that way. It is a duration in minutes (1–180), so analyze it as a numeric metric, not a date or time. The dataset has no event date, so you can't do time-trend analysis. - **Pipeline columns:** `_batch_id`, `_source_file` and `_source_system` each have a single constant value. They are metadata with no analytical use and should be excluded from modeling. - **`bedtime_phone_minutes` privacy:** the card flags it as person-identifying, so report it only in aggregate.