Study S5 · August 2026
Does last night’s sleep score predict today’s run?
Key finding
Across 7,492 steady runs, the median within-athlete correlation between last night's Garmin sleep score and today's running efficiency was 0.04. The same pipeline found 0.437 on a control pair, so the method works — the relationship simply is not there.
Gneta, August 2026 · our athletes · 7,492 paired runs
54
athletes
7,492
paired steady runs
0.04
median correlation
What we did
We joined each night's sleep score to the run the athlete did the following day, keeping only steady efforts. That gave 7,492 pairs across 54 athletes with at least 20 pairs each. A daily record dated D carries the sleep that ended on the morning of D, so the join really is last night's sleep against today's run.
“Efficiency” here is speed divided by average heart rate on steady runs — 20 to 90 minutes, average HR between 120 and 175, over 3 km — then z-scored within each athlete so people are only ever compared against themselves. It is a rough proxy for how good a run felt in physiological terms, not a lab measurement.
What we found
Finding 1
Last night’s sleep score tells you almost nothing about today’s run
Median within-athlete correlation: 0.04, middle half between −0.029 and 0.15. Only 3.7% of athletes reached 0.3.
70.4% of athletes were positive, which is a faint lean in the direction common sense expects — but nowhere near strong enough to plan a session around.
Finding 2
Sorting by sleep quintile barely separates the runs
Worst-sleep quintile, efficiency z-scored within athlete: median −0.035. Best-sleep quintile: +0.079. The three middle quintiles sit between them within a hundredth or two of each other.
That is roughly a ninth of a standard deviation between your worst nights and your best — swamped by terrain, weather and how hard you decided to go.
Finding 3
The control came back strong, so the null stands
Sleep score against Training Readiness — two numbers that genuinely share inputs — produced a median of 0.437 across the cohort. 100% of athletes were positive and 89.1% cleared 0.3.
Same athletes, same code, same window. When a real relationship was present, this method found it comfortably. Against actual running performance, it found 0.04.
Limits you should know before citing this
- The cohort is self-selected. Garmin owners who sought out a third-party analytics tool. Fitter and more data-curious than average, and not a random sample of anyone.
- One night is not sleep debt. We tested last night against today. Accumulated deprivation over a week or a month is a different question and this study does not touch it.
- Efficiency is a proxy, not a verdict on the run. Speed per heartbeat cannot see how a run felt, and it is confounded by terrain, heat and pacing choices.
- Behaviour absorbs some of the effect. Athletes who sleep badly and then run easier — or skip the run entirely, which removes the pair from the data — flatten exactly the relationship we are looking for.
- Six accounts excluded. Near-identical multi-year histories under different Garmin accounts, created inside a known incident window. See the methodology.
- Observational. No causal claim is made anywhere in this study.
The data
The aggregate output and the script that produced it are public. Every number on this page traces to that file.
View the aggregate data (JSON) →