**Overall, the data is structurally clean. The one real problem is that `sleep_debt_category` does not follow cleanly from `total_sleep_hours`.** **What is clean** - **Keys and nulls:** There are 8,500 rows and 8,500 distinct `user_id`s, so no duplicates. Every measure I profiled has 0 nulls (step-0, step-5). - **Impossible values:** None found. All ranges are physically plausible (step-0, step-5): | Column | Range | |---|---| | `age` | 18–65 | | `bedtime_phone_minutes` | 1–180 | | `screen_brightness_pct` | 10–100 | | `blue_light_filter_active` | 0 or 1 | | `caffeine_post_5pm_mg` | 0–250 | | `physical_activity_min` | 0–112 | | `total_sleep_hours` | 3.2–9.8 | | `morning_alarm_snoozes` | 0–7 | | `next_day_fatigue_score` | 1–10 | `deep_sleep_pct` and `rem_sleep_pct` never sum past 100 (`deep_rem_over100` = 0). - **Formats and units:** The categorical columns have no spelling, case or whitespace variants. Chronotype has 3 values, gender 3, occupation 5, bedtime app 6, and sleep debt category 4 (step-1). The numeric columns are typed as numbers, and I found no sign of mixed units. **Issues and oddities** 1. **Category does not match its source measure.** The card says `sleep_debt_category` is derived from `total_sleep_hours`, but the hour ranges of two categories overlap (step-3, step-8): - Mild Deficit (6.75–7.74 h) and Optimal Recovery (7.75–9.8 h) have clean cut-offs. - Severe Sleep Debt spans 3.2–5.38 h and Moderate Debt spans 3.2–6.74 h, so the two overlap across 3.2–5.38 h. - 207 Moderate Debt rows sleep under 4 hours, while only 440 of the 647 Severe rows do (step-4, step-7). - Severe therefore includes people sleeping up to 5.38 h, while some Moderate users sleep as little as 3.2 h. The label likely uses another input, such as latency, or an undocumented rule. Don't treat it as a pure function of hours, and don't use it as a target alongside `total_sleep_hours` without checking this. 2. **Long sleep-latency tail.** `sleep_latency_min` runs up to 123.3 minutes. 104 Severe rows and 13 Moderate rows exceed 90 minutes (step-2). These are plausible for severe insomnia but are extreme outliers relative to the overall median of 37.6 minutes (step-5). Verify them before using the column in averages. 3. **Possible floors and caps.** Several columns end at round or hard limits: minimum latency of exactly 6.0, minimum sleep of 3.2, and brightness from 10 to 100. This may reflect clipped or generated data. I didn't check the distributions at the edges, so treat it as a hint rather than a finding. 4. **Zero-inflated caffeine.** 4,950 of 8,500 rows (about 58%) have 0 mg of post-5pm caffeine (step-5). This is plausible, but a zero-heavy column needs care in correlations and averages. 5. **Small group.** Non-Binary has only 248 rows (step-1), so group-level statistics for it will be noisy. 6. **Sensitive column.** `bedtime_phone_minutes` is flagged as person-identifying, so report it only in aggregate. I did not run per-column outlier tests such as IQR or z-scores beyond min, max and median, so more subtle outliers inside the valid ranges may exist.