I Tracked 400 Nights of Sleep. Five Things Moved the Needle

I have worn a tracker every night since 2019. In 2024 I stopped browsing the graphs and started running the thing properly: one variable at a time, two-week blocks, alternating with control blocks, decided in advance and written down before I started so I could not talk myself into a result afterwards.

Four hundred nights. Fourteen interventions. Here is what held up.

The honest caveats first

This is an n-of-1. It is unblinded — I knew which block I was in, and expectation affects both sleep and how you rate it. Consumer trackers estimate stages from movement and heart rate rather than measuring brain activity, so their stage numbers should be treated as a rough index, not a measurement. I leaned on total sleep time, sleep efficiency and resting heart rate, which wearables handle far better than staging.

Treat the following as a set of leads worth testing on yourself, not as findings.

What worked

1. A fixed wake time, including weekends

The largest single effect, and the one I least wanted to be true. Holding wake time within a 30-minute band every day, including Saturday and Sunday, cut my sleep-onset latency roughly in half over a fortnight and reduced night-to-night variance more than anything else I tried.

The cost is real: no lie-ins. I have accepted this, resentfully.

2. Dropping the bedroom to about 18°C

Core temperature has to fall for sleep to consolidate, and a warm room fights that. Moving from a habitual 21°C to 18°C produced a clear reduction in awakenings after sleep onset. This is one of the few interventions where the mechanism, the lab literature and my own data all point the same direction.

3. Caffeine cut-off at 2pm

I had assumed I was a fast metaboliser because coffee at 6pm did not stop me falling asleep. Falling asleep was never the problem. Moving the cut-off from 5pm to 2pm did not change my onset latency at all — it changed sleep efficiency in the back half of the night, which I would never have noticed without the log.

4. Ten minutes of outdoor light before 9am

Cheap, boring, effective. The effect showed up as a gradual phase shift over about ten days rather than an overnight change, which is exactly what you would expect from circadian entrainment and exactly why a three-day trial would have told me it was useless.

5. Getting out of bed when awake at 3am

Counterintuitive, and the one that took most discipline. Lying in bed awake trains an association between the bed and wakefulness. Leaving the room and returning when sleepy is the core of CBT-I, the first-line treatment for insomnia. It made my worst nights noticeably less bad, though it did nothing to my averages.

What did nothing measurable

• Magnesium glycinate. Two separate blocks. Nothing outside noise. Plenty of people report otherwise and the safety profile is benign, so I am not telling anyone to stop — I am reporting that I could not find an effect.
• Blue-light blocking glasses in the evening. No detectable change in onset or efficiency. Dimming the lights entirely did have an effect, which suggests intensity mattered more than spectrum for me.
• A white noise machine. Neutral on average, mildly negative on a few nights.
• Sleeping with a weighted blanket. Pleasant. Inert.
• Mouth taping. Abandoned after four nights. Unpleasant, and if you suspect disordered breathing this is a question for a clinician and a proper study, not for tape.

What made things worse

Alcohol, unambiguously, and worse than I expected. Two drinks with dinner reliably knocked several points off sleep efficiency and raised my overnight heart rate by a margin that dwarfed every positive intervention on this list. Alcohol shortens sleep latency and wrecks the architecture behind it — the sedation is not sleep.

Late high-intensity training was second. Anything hard after 8pm cost me the first hour of the night.

What I would tell someone starting out

Run fewer things for longer. My first year of tracking was worthless because I changed three variables at once and read the graphs like tea leaves. The moment I switched to two-week blocks with a pre-registered plan, four hundred nights produced five useful facts — which is a low yield, and still more than I got from five years of vibes.

Not medical advice. If your sleep is genuinely broken rather than merely suboptimal, CBT-I delivered by a clinician has a far better evidence base than anything in this article.