Garmin Metrics
Does Your Garmin Sleep Score Predict Tomorrow's Run?
August 1, 2026
Does your Garmin sleep score predict how you'll train tomorrow?
Almost not at all. Across 7,492 steady runs matched to the sleep score from the night before, the median athlete's correlation between sleep score and running efficiency was 0.04, which is close enough to zero that no runner should plan around it.
That is not the answer I expected to write, and it is not the answer the rest of the internet gives you. So before anything else, here is the check that makes me believe it.
The positive control, first
The obvious objection to a null result is that the measurement is broken. Wrong dates, wrong join, wrong metric, and you get noise no matter what is true underneath.
So I ran the same pipeline against something that has to correlate. Garmin's Training Readiness score takes sleep as a direct input, by construction. If our join and our method work, sleep score and same-day Training Readiness should move together, hard.
They do. Median within-athlete correlation: 0.437. Every single athlete in that group came out positive, and 89.1% cleared 0.3. That is what a real relationship looks like in this dataset, computed on the same dates, the same accounts, and the same code.
Same machinery, two outcomes. Sleep score predicts the number Garmin derives from sleep. It does not predict the run.
How we set this up
Our platform holds Garmin daily summaries and activity files for athletes who connected their accounts. For this study (S5, computed 2026-08-01) I pulled every steady run that had a sleep score on the matching date: 7,492 runs across 54 athletes with at least 20 paired runs each. Six accounts were excluded because they were created during a May 2026 incident window with near-identical multi-year histories under different Garmin logins, so they cannot be treated as independent athletes.
One detail worth stating openly, because it decides what the study means. Garmin's daily summary for a given date carries the sleep that ended that morning. A same-day join is therefore last night's sleep against today's run, which is exactly the question people actually ask. If that convention were inverted, the readiness control above would not have come out at 0.437.
The performance outcome is running efficiency: speed divided by heart rate, then z-scored within each athlete so everyone is compared only against their own normal. Steady runs only, because comparing an interval session to a recovery jog measures the plan, not the athlete.
Standard caveat, and it applies to every number here: these are self-selected Garmin owners who connected a training-analytics app. Fitter and more data-curious than the general population, near-certainly.
What 0.04 looks like
The spread across athletes is narrow and centred on nothing. Median rho 0.04, quartiles at -0.029 and 0.15. Only 3.7% of athletes cleared 0.3, and not one fell below -0.3.
There is one grain of signal worth naming: 70.4% of athletes were weakly positive. Better sleep nudging better efficiency shows up more often than chance would give you. It is a consistent direction with a trivial magnitude, and those are different things.
The quintile view says the same thing in units you can feel. Sorting each athlete's runs into five buckets by their own sleep score:
| Sleep quintile | Median efficiency (z) | Runs |
|---|---|---|
| q1 (worst sleep) | -0.035 | 1,508 |
| q2 | +0.044 | 1,450 |
| q3 | +0.047 | 1,471 |
| q4 | +0.095 | 1,468 |
| q5 (best sleep) | +0.079 | 1,434 |
The ordering is roughly right, and it does not reverse. But the gap from your worst nights to your best is about 0.1 standard deviations of your own efficiency, and the top quintile is not even the peak. For a runner whose steady pace varies by 15 seconds per kilometre week to week, that is a second or two, buried under a spread that runs from roughly -1.3 to +1.1 z inside every single quintile. Your worst-sleep runs and your best-sleep runs overlap almost completely.
Two readings, and I will not pretend to settle it
Reading one: one night matters less than the internet says. Chronic sleep debt has real, well-documented effects on athletes. A single 61-instead-of-84 night, on a body that is otherwise fed, trained and rested, may simply not move an aerobic 45-minute run. The physiology that carries a steady run is not that fragile over 24 hours.
Reading two: the outcome is too noisy to catch a modest effect. Daily running efficiency absorbs terrain, heat, humidity, wind, shoes, fuelling, how much of the run was into a headwind, and how hard you decided to go. If sleep moves efficiency by 0.05 z and the day-to-day noise is 1.0 z, no correlation over 7,492 runs will find it cleanly, and a real effect stays hidden.
I lean toward the first, mildly, because 70.4% of athletes pointing the same way suggests the method is not blind. But 7,492 runs cannot separate these two stories, and anyone who tells you they can from this kind of data is overselling. What the data does rule out is the strong version of the claim: that your morning sleep score is a usable forecast of today's run. It isn't.
Where we got this wrong before
Our own guide to using Garmin sleep data as an athlete ends with a line about noticing that your Thursday interval quality drops in the weeks your Monday sleep score is below 60. That framing implies a night-to-night link tight enough to see in your own data. On our numbers, most athletes will not see it, and the ones who think they do are probably reading noise. The rest of that article holds up. That sentence oversold what a single night's score can tell you, and I would not write it again.
Four things this study cannot see
- Efficiency is confounded. Speed over heart rate is dragged around by hills, heat, pacing and intent. Z-scoring within athlete removes their average conditions, not any given day's.
- Sleep score is a composite, not sleep architecture. Garmin's 0-100 number is duration, quality, deep and REM blended into one figure, and wrist-based sleep staging is known to be imprecise. A null against the composite is not a null against sleep.
- Acute is not chronic. This tests last night against today. It says nothing about what six weeks of five-hour nights does to you, which is the version of the question that probably matters most.
- Steady runs only, self-selected cohort. No races, no intervals, no maximal efforts. It is entirely possible sleep bites hardest exactly where we did not look.
What to do with your sleep score instead
Stop treating the morning number as a forecast, and start treating it as a log entry.
Read it in weeks, not nights. A run of poor scores is worth acting on. One bad night is not a reason to downgrade a session you were ready for yesterday.
Use the recovery metrics that were built to be read daily. HRV status is a rolling baseline rather than a single-night verdict, which is exactly why it behaves better as a decision input, and our explainer on how Garmin computes HRV status covers how to read the colour bands without overreacting. If you want an early illness signal, overnight respiration rate is steadier still.
Know what Training Readiness is telling you. At 0.437, readiness is substantially a restatement of last night's sleep. That makes it a good summary of how recovered you are, and a weak forecast of how the run will go, for the same reason the sleep score is. The same applies to Body Battery, which inherits the overnight signal too.
Let the run tell you. Ten minutes in, your heart rate at your usual easy pace is a direct measurement, not an inference from wrist accelerometry. It has never once been wrong about what your body is doing right now.
Frequently Asked Questions
Is my Garmin sleep score accurate?
This study did not test that, and it is a different question. We tested whether the score predicts next-day running efficiency, not whether it measures your sleep correctly. Worth knowing that wrist-based sleep staging is imprecise, so the underlying inputs to the composite carry error of their own.
Should I skip a workout after a bad Garmin sleep score?
Not on the score alone. In 7,492 runs, athletes in their worst sleep quintile ran about 0.1 standard deviations below their best-quintile efficiency, with almost total overlap between the groups. If you feel wrecked, that is information. The number by itself is not.
Why does my Training Readiness drop when I sleep badly, if it doesn't affect my run?
Because sleep is a direct input to readiness. We measured that link at rho 0.437 with 100% of athletes positive. Readiness is faithfully reporting that you slept badly. Our data suggests it is not thereby reporting how your run will go.
Does sleep matter for performance at all, then?
Chronic sleep loss has a substantial evidence base behind it in sports science. Our study only tested the acute, one-night version against steady aerobic runs. Those are not the same claim, and nothing here argues against protecting your sleep over a training block.
How does this compare to other Garmin metrics you've tested?
The pattern keeps repeating: metrics measured against the wrong benchmark look broken or useless. Our study of the race predictor's accuracy found it missed training efforts by 22.7% while landing near-exactly on real races. Sleep score is the mirror image, a number that describes something real about last night while telling you little about the next 24 hours.
The uncomfortable part of this result is that a bad night genuinely feels bad. Your watch agrees, your stress numbers agree, and your readiness score drops right on cue. Then you go out and run your normal pace at your normal heart rate, 7,492 times over. Whatever the sleep score is measuring, and it is measuring something, it stops short of the road. Check your sleep data for the trend. Do not let it pick your workout.
