**The most counterintuitive finding is that the usual sleep-hygiene advice barely shows up in this data, while bedtime phone minutes dominate.** These are Pearson correlations across all 8,500 users (the step-0 query): | Behavior | Outcome | Correlation | |---|---|---| | Bedtime phone minutes | Sleep latency | **0.86** | | Bedtime phone minutes | Next-day fatigue | **0.71** | | Bedtime phone minutes | Total sleep hours | **-0.55** | | Blue-light filter on | Sleep latency | -0.057 | | Blue-light filter on | Sleep hours | +0.039 | | Caffeine after 5pm | Sleep hours | -0.038 | | Caffeine after 5pm | Deep sleep % | -0.34 | | Physical activity minutes | Sleep hours | 0.004 | | Physical activity minutes | Fatigue | -0.029 | | Screen brightness | Sleep latency | 0.093 | - **Blue-light filters do almost nothing.** Users with a filter on fall asleep only marginally faster and sleep no longer. The correlations are close to zero, so the filter looks close to irrelevant next to how long people use their phones. - **Late caffeine doesn't shorten sleep.** Its correlation with total sleep hours is -0.038. It does show some link to a longer time to fall asleep (0.20) and to less deep sleep (-0.34), so caffeine seems to affect sleep quality more than sleep quantity. - **Exercise and screen brightness matter little.** Activity is uncorrelated with sleep hours and fatigue, and brightness is only weakly tied to latency (0.093). - **Phone minutes, not "what" you do on the phone, drive the problem.** A correlation of 0.86 between phone minutes and sleep latency is very strong for behavioral data. It chains into the other outcomes: sleep latency is correlated at -0.53 with sleep hours, and sleep hours at -0.88 with fatigue. I'd treat the near-zero filter, caffeine and activity results with caution. These are linear correlations on observational data, so a nonlinear or subgroup effect could be hidden, and I haven't tested for one. Still, if the aim is to cut sleep latency and fatigue, shortening bedtime phone use looks far more promising than recommending filters, a caffeine cutoff or more exercise.