Sleep is the cheap performance lever

May 25, 2026

The cleanest correlation I have in two years of training data is between mean nightly sleep and FTP at the next 8-week test. Not nutrition. Not Z2 volume. Sleep.

I track it with an Oura ring. The metric that matters isn’t total sleep time, which is contaminated by time in bed scrolling. It’s the deep-plus-REM total, which on adults sits around 25-30% of total sleep when things are working. Mine, averaged across the last 90 days: 113 minutes deep + REM per night, on an average of 7h 24m total. Ratio 25.5%, low end of normal.

The interventions that moved the number, ranked by effect size:

Caffeine cutoff at noon. Caffeine has a half-life of about 5-7 hours. An espresso at 4pm leaves you with a quarter-dose still circulating at 10pm. The night after a noon cutoff, deep sleep gains 15-20 minutes on average for me. The night after a 4pm coffee, it loses about the same.
Bedroom at 18°C. Core temperature drop is the gating signal for sleep onset and for cycling into deep. Above 20°C I lose roughly 12 minutes of deep over the night.
No food within 3 hours of bed. Digestion suppresses HRV and pulls cardiac output away from the recovery profile. Recovery scores correlate with empty-stomach bedtime more than with food quality.
Morning sunlight within 30 minutes of waking. The circadian gain shows up the next night, not the same one. Hard to feel in the moment, real in the 30-day average.
Magnesium glycinate, 400 mg, an hour before bed. Marginal but consistent. Roughly 5 minutes of additional deep sleep per night, holding everything else constant. Not life-changing. Worth the dollar.

The interventions that did nothing measurable for me, despite popular claims:

Blue-light glasses after sunset.
Mouth tape (I don’t snore, so this was always a long shot).
L-theanine.
Tart cherry juice.

The cheapest performance lever in endurance sport is going to bed at 10pm instead of 11pm for ninety consecutive days. Most of us don’t, and the reason isn’t that we don’t know.