What your wearable can and cannot tell you about your energy

Someone shows me a green recovery score and tells me they feel awful. They are not confused, and their ring is not broken. They are describing two different things.

Ask whether fitness trackers are accurate and you will get an answer about step counts and heart rate. The more useful question is what they are accurate about. These devices measure the body at the wrist or the finger. How energetic you feel is produced in the brain, and no consumer device measures that directly. The only way to know it is still to ask.

It is also not one thing. In our clinical model, the felt sense of energy has five partly separable components: how awake and sharp you are, how much drive you have, how calm or on edge you feel, and your mood. They come apart in practice. Chervin studied 190 people with obstructive sleep apnoea and asked them to name their single most significant symptom. About 40% chose lack of energy and about 22% chose sleepiness, and objective measures of sleepiness and apnoea severity showed little or no association with any of the symptoms [1].

Your wearable picks up body signals linked to the first of these. The other three, it cannot see.

Three assumptions worth reconsidering

That a good sleep score means you slept well

The assumption: the score is a verdict on your night.

What the device is actually doing: an accelerometer records movement, and an optical sensor shines light into the skin and reads the pulse that comes back. From those two channels an algorithm infers when you fell asleep, how long you stayed there, and which stage you were in. Rings carry no gyroscope at all and stage sleep about as well as watches do, which tells you the motion channel that matters is the accelerometer [2].

In one of the most thorough comparisons against laboratory sleep studies (polysomnography), Chinoy tested seven consumer devices over three nights each. Every device detected sleep at least 93% of the time, but correctly identified time awake only 18% to 54% of the time [3]. These devices are very good at agreeing that you are asleep and poor at noticing that you are not. Stage estimates carried biases of up to 36 minutes, and the most common error was the same across devices: wake, deep and REM sleep all get swept into "light sleep".

The better question: what changed on the nights that felt different? Kaplan gave 1,483 older adults a night of home polysomnography, fed 38 sleep and clinical variables into machine learning models, and asked them to predict how good people said their sleep was. The models explained 11% to 17% of the variance, and the amount of slow wave sleep was not among the important predictors [4]. If a full sleep laboratory cannot tell you how your night felt, a ring will not either.

That a recovery score tells you how much energy you have

The assumption: readiness and energy are the same measurement.

Readiness and recovery scores are composites, mostly of sleep duration, sleep quality estimates and heart rate variability. Heart rate variability is the variation in the time between consecutive heartbeats, and RMSSD, the measure most devices report, is a reasonable measure of cardiac vagal tone, the calming input to the heart [5].

What it is not is a stress dial. The popular "sympathetic versus parasympathetic balance" reading comes from the LF/HF ratio, and Billman's paper on it is titled, without hedging, The LF/HF ratio does not accurately measure cardiac sympatho-vagal balance [6]. Heart rate variability is sensitive but not specific. Across 28,175 people, Altini and Plews found it fell about 12% after heavy drinking and about 10% during illness, shifted about 5% with training, and changed across the menstrual cycle [7]. One number, many causes.

The better question: what is the score not able to see? Drive, calm and mood leave no peripheral signature a wearable can read. If you want to understand your energy, you have to record your energy.

That more data means more understanding

The assumption: another metric will finally explain the pattern.

A metric only helps if it captures something that causes the outcome, and a correlation inside your own data cannot tell you which way the arrow points.

Take a pattern I see often: years of tracker data, and no explanation for why some nights leave someone flat. The last coffee of the day is at about 4pm. Pooled across 24 studies, caffeine shortens total sleep time by about 45 minutes and cuts deep sleep by about 11 minutes, and Gardiner's dose-scaled estimate puts the cut-off for a single 250 ml coffee at roughly 8.8 hours before bed [8]. For an 11pm bedtime that is early afternoon. The finer effect is on slow wave activity, the brainwave (EEG) signal that shows how deep the sleep actually is. Landolt showed that 100 mg at bedtime, one modest cup, suppresses it through the first sleep cycle [9].

A wearable is unlikely to have found this, because deep sleep is among the things these algorithms estimate least well [3]. What finds it is recording how you feel, forming a hypothesis, and testing it by moving the coffee.

The better question: what would have to be true for this to be the cause?

Where fitness trackers are accurate, and genuinely useful

Movement and heart rate. Germini's review of 65 studies found that the best-performing devices counted steps to within about 25% and measured wrist heart rate to within about 10%, but accuracy varied widely between brands. Energy expenditure was off by more than 30% for every brand tested, and the authors concluded that none of the devices measured calories accurately [10]. Treat that number as a rough guide at best.

Overnight heart rate. This is the signal I trust most, because it is measured rather than inferred, and it answers to what you did the evening before. In 4,098 Finnish employees, Pietilä found heart rate variability suppressed through the first 3 hours of sleep after alcohol, dose-dependently and even at low intake [11], and Strüven measured about 3 extra beats per minute across 3 drinking days [12].

A heavy late dinner does something comparable. Tada randomised 24 women to a large or a small evening meal with the same daily energy intake and found parasympathetic activity reduced across the night, most clearly in the first 3 hours [13]. In a 7.5-week randomised trial, Grimaldi had one group finish eating at least 3 hours before bed and measured night-time heart rate about 2 beats per minute lower, with the overnight dip about 5 percentage points deeper [14].

Consistency over weeks. In 60,977 people wearing research accelerometers, Windred found sleep regularity predicted mortality more strongly than sleep duration, with a fully adjusted hazard ratio of 0.70 for the most regular fifth against the least. The most regular fifth fell asleep and woke within roughly 1-hour windows; the least regular, within roughly 3-hour windows [15].

A prompt to look closer. Watch-based sleep apnoea features now carry US regulatory clearance. In the data submitted for one, the feature correctly flagged about 83% of people with moderate to severe sleep apnoea and correctly cleared about 88% of those without it [16]. That matters, because most moderate to severe sleep apnoea has been found to go undiagnosed [17].

What to do when the score and the feeling disagree

Start by treating the feeling as a measurement, not an error. It is the outcome you actually care about, and it is the only access anyone has to three of the four components of energy.

Be precise about what a raised overnight heart rate means, though. In the meal-timing studies, sleep stages were unchanged [13, 14, 18], and the same holds for evening training: in a small study of trained runners, a hard session raised overnight heart rate by about 3 beats per minute while total sleep time went up and heart rate variability was unaffected [19]. A higher trace is not evidence your sleep stages were wrecked. It is evidence your body was still working when it should have been winding down, which is a more useful thing to know.

The dissociation runs both ways. Van Dongen restricted 48 adults to 4, 6 or 8 hours in bed for 14 nights. Those on 6 hours accumulated cognitive deficits equivalent to up to 2 nights of total sleep deprivation, while their own sleepiness ratings did not distinguish 6 hours from 4 [20]. Feeling misses things too. Neither quantity substitutes for the other.

So record both, with the context. On the days the two diverge, write down what was different: the drink, the late meal, the training session, the last coffee.

Then look at what a wearable cannot see at all. It knows nothing about your thyroid, your iron stores or your inflammatory tone. Houston's review of 18 trials found that iron improved felt fatigue modestly in non-anaemic adults with iron deficiency, while leaving measured exercise capacity unchanged [21]. What changed was how people felt, not what their bodies could measurably do.

And set expectations honestly. In primary care, depression is the commonest identifiable cause of tiredness at about 18.5%, while serious physical disease is found in about 4.3%, no more often than in people without the symptom [22]. Persistent low energy is worth clinical attention. A broader panel on its own is not the answer.

Where Calibre fits

The problem with a proxy measurement is not that it is wrong. It is that there is nothing to interpret it against.

We start from the outcome. Our energy score asks about all four components (wakefulness, drive, calm and mood) alongside the context of what made a good day and a bad one, because that is the only thing in the whole system that measures the experience rather than a correlate of it.

Around that we read your wearable trends together with diagnostics and your clinical history, so the pattern has something to be explained by. Our causal model then tests which factors are plausibly driving your pattern rather than simply moving alongside it, and shows where your evidence has gaps.

And a clinician tells you which of your metrics matter and which do not. Calibre is clinician-led, with a CQC-regulated clinical partner.

FAQs

Are fitness trackers accurate enough to base health decisions on?

For step trends, overnight heart rate and patterns over weeks, broadly yes. For calories burned and deep sleep minutes, no [3, 10]. They can flag a possible problem, not diagnose one.

Why do I have a good sleep score but still feel tired?

Because the score estimates your night, and how energetic you feel is a brain state. Even a full sleep laboratory explains only 11% to 17% of how good people say their sleep was [4].

What does a readiness or recovery score actually measure?

A composite of estimated sleep and heart rate variability. Heart rate variability shifts with training, alcohol, illness and the menstrual cycle, so it flags that something changed, not what [7].

Which sleep tracker is the most accurate?

A 2026 meta-analysis found no device consistently superior; all overestimated sleep and underestimated time awake [23]. How many nights you actually wear it matters more than the brand.

Should I stop using my wearable if the scores do not match how I feel?

No. Keep it for duration, regularity and overnight heart rate, and stop reading the composite score as a verdict.

Can a wearable pick up something a blood test would miss?

Yes. No blood test shows how you slept, or how consistently.

How long do I need to wear a tracker before the data is useful?

Heart rate variability is individual, so it only means anything against your own baseline. Judge weekly averages rather than single days. In one study, a weekly average built from as few as 3 valid readings tracked the full week closely [24].