**The most counterintuitive finding is that late caffeine barely changes how long people sleep, but it strongly cuts deep sleep.** Across all 8,500 users, post-5pm caffeine has these correlations (step-7 correlation query): | Outcome | Correlation with caffeine after 5pm | |---|---| | Total sleep hours | -0.04 | | Deep sleep % | -0.34 | | Sleep latency | +0.20 | | Next-day fatigue | +0.18 | - **Duration versus quality:** The usual advice ("caffeine cuts your sleep") would predict a clear drop in hours. Instead, duration is essentially unaffected, and the damage shows up in sleep quality. Deep sleep is the most strongly caffeine-linked sleep-architecture measure here. - **Phone use dwarfs it:** Bedtime phone minutes correlate +0.86 with sleep latency, +0.71 with fatigue and -0.55 with sleep hours. Caffeine is a minor factor next to that. - **Fatigue tracks duration:** Fatigue is tied to short sleep (-0.88 with sleep hours) and to alarm snoozes (+0.88). Snoozing looks like a symptom of short sleep, not a cause. **Two other findings that go against expectation:** - **Exercise shows no link to sleep:** Physical activity has near-zero correlation with sleep hours (+0.004) and with fatigue (-0.03). - **Blue-light filters do little:** Filter users and non-users differ only slightly (step-3 grouped averages). | | Filter on (3,976 users) | Filter off (4,524 users) | |---|---|---| | Average sleep latency | 39.6 min | 41.6 min | | Average sleep | 6.33 h | 6.21 h | The correlations with latency (-0.06) and sleep hours (+0.04) are tiny. Screen brightness is also only weakly related to latency (+0.09). Phone time is what matters, not how the screen is configured. **Caveat:** These are observational correlations. They don't show causation, and I did not control for confounders such as occupation, chronotype or age.