Here is a representative sample of the dataset. It comes from the sample queries already run and rendered above, which excluded `bedtime_phone_minutes` because it is flagged as person-identifying. The dataset has 8,500 rows, one per `user_id`, with 22 columns. **How the samples were drawn** - **Random draw:** two random draws of about 15 rows each. Both show the same broad mix of genders, occupations, chronotypes and bedtime apps. - **Stratified draw:** 4 rows from each `sleep_debt_category`, because the categories are very unevenly sized. A plain random sample would show very few Severe cases. **Category mix** (from the category breakdown above) | Category | Rows | Share | |---|---|---| | Moderate Debt | 4,462 | 52.5% | | Mild Deficit | 2,004 | 23.6% | | Optimal Recovery | 1,387 | 16.3% | | Severe Sleep Debt | 647 | 7.6% | **What the rows show** - **Category follows sleep hours.** `sleep_debt_category` tracks `total_sleep_hours` closely. Severe rows sleep about 3.8 hours on average, Moderate about 5.6, Mild about 7.2 and Optimal about 8.3. - **Worse sleep goes with worse mornings.** Average `next_day_fatigue_score` rises from 1.08 (Optimal) to 9.58 (Severe). Average `morning_alarm_snoozes` rises from 0.81 to 5.88. - **Sleep latency also climbs.** Average `sleep_latency_min` is about 29 minutes for Optimal and about 73 for Severe. - **Fields that look the same across categories.** Age, brightness, caffeine and activity look similar across categories. Average age is about 33–35 in every category, and average screen brightness is 54–61%. - **Sparse columns.** Many rows have `caffeine_post_5pm_mg` = 0, and the median is 0 in three of the four categories. Many also have `physical_activity_min` = 0. - **Categorical values.** `blue_light_filter_active` is coded 0/1. Gender includes Female, Male and Non-Binary. Chronotype is Morning Lark, Intermediate or Night Owl. Occupations seen include Corporate 9-to-5, Healthcare / Shift Worker, Remote Tech, Freelance / Creative and Student. Apps include TikTok / Reels, YouTube, Instagram / Reddit, Messaging / Chat, Streaming and News / Reading. I can break the sample down by chronotype, occupation or app if you want a closer look.