The sample rows are in the results above: a 15-row sample, a 16-row sample with 4 users from each sleep-debt category, and a 15-row sample. Nothing new needed to be run. **What the samples show** - Each row is one user (`USR-xxxxx`), with demographics (age, gender, occupation, chronotype), bedtime habits (primary app, screen brightness, blue-light filter on or off, caffeine after 5pm, physical activity) and sleep outcomes (latency, total hours, deep and REM %, alarm snoozes, next-day fatigue, sleep-debt category). - The 16-row stratified sample takes 4 users from each category, so it covers Optimal Recovery, Mild Deficit, Moderate Debt and Severe Sleep Debt equally. The real mix is skewed: Moderate Debt is 52.5% of the 8,500 users (4,462), Mild Deficit 23.6% (2,004), Optimal Recovery 16.3% (1,387) and Severe Sleep Debt 7.6% (647). Use the other two samples, not the stratified one, to see how common each type is. - The samples include a spread of occupations (Corporate 9-to-5, Healthcare / Shift Worker, Remote Tech, Student, Freelance / Creative), all three chronotypes, all six bedtime apps and all three genders (Female, Male, Non-Binary). **Patterns visible in the samples** - The sleep-debt categories line up with sleep hours. In the full data the averages are 8.35 h for Optimal Recovery, 7.23 h for Mild Deficit, 5.55 h for Moderate Debt and 3.76 h for Severe Sleep Debt. - Worse categories also have more morning snoozes (0.81 → 1.81 → 3.35 → 5.88 on average), higher fatigue (1.08 → 1.83 → 4.68 → 9.58) and longer sleep latency (28.9 → 34.7 → 42.4 → 72.7 minutes). - Severe Sleep Debt rows in the sample have fatigue scores of 8.8–10 and latencies of 60–84 minutes, well above the other categories. **Caveats** - Some values are zero: many users have 0 mg of caffeine after 5pm (the median is 0 in three of the four categories) and a few have 0 minutes of physical activity. These are real values, not obviously missing data. - I left `bedtime_phone_minutes` out of the samples because it is flagged as person-identifying. I can aggregate it by segment if you want.