To judge smart ring activity tracking accuracy, first separate three things that a companion app often shows side by side: a step count, an activity estimate (movement intensity turned into calories and active time), and records that other apps write into a shared health hub. They come from different processes and are validated differently. This article covers activity records only. It does not cover recovery scores or sleep interpretation.
Step display vs. activity estimate
A step count answers "how many steps did the algorithm detect?" From finger or wrist motion, the device infers step-like patterns. Manufacturers describe their own methods, but a description of a method is not an independent test.
Activity estimates are a different layer. Intensity is often expressed in METs, where one MET is resting metabolic rate. Active calories, activity time and any "walking equivalent" distance are modeled from motion, body metrics (such as age, sex, height and weight) and thresholds that each vendor chooses. A walking-equivalent distance, where offered, is usually a way of expressing energy, not mileage actually traveled. Check your app's own definition before using it.
So don't use one measure to audit the other. A day with a modest step count and a hard strength session can look very different across steps and energy estimates, and neither is necessarily wrong.
What the independent validation covers
The most relevant primary study for a ring is the 2023 validation of ring-measured energy expenditure and steps in laboratory and free-living conditions, part of the MEDAL project in southwest Sweden. It tested one research ring, a second-generation model running firmware 4.0.4. Its design:
- Laboratory part: structured activities with the ring's MET output compared against indirect calorimetry (oxygen consumption), with accelerometers on the hip, thigh and wrist.
- Free-living part: 32 participants wore the ring, a pedometer and the three accelerometers for 14 days. Ring step count was compared with the reference monitors, as were MET, total and active energy expenditure.
- Self-report: participants kept daily logbooks of activity, sleep times and device removal, which the authors used to interpret possible errors.
Reported results for that ring and firmware:
- Correlations were strong: r = 0.93 against indirect calorimetry in the lab, and r ≥ 0.76 against reference monitors for all free-living variables.
- Correlation is not agreement. The study also found significant mean differences, including a free-living step-count difference of 2124 ± 4256 steps and total energy expenditure differences of 362–494 kcal.
- In the lab, the ring tended to underestimate energy expenditure, and the gap grew with intensity: −0.12 MET sitting, −0.82 MET fast walking and −3.49 MET very fast running.
- The authors concluded that the ring could not unambiguously be recommended as interchangeable with the reference monitors, while high correlations suggest it can show group-level differences.
Two limits matter. First, these findings belong to the tested ring and firmware. Sensors, firmware and algorithms change between generations and software versions, so treat the figures as evidence about that setup, not as error rates for any ring on your finger today, and not for a wrist band. Second, 32 free-living participants in one region, with different reference devices for different outputs, supports "strong association with systematic differences", not a single accuracy percentage for steps.
A systematic review of commercially available wearables offers broader context on steps, energy expenditure and heart rate across devices. Use it for general expectations about consumer-tracker validity, not for the performance of any single device.
The same logic applies to your own comparison: another device's step count is not ground truth. The study used several reference monitors, not one, as its yardstick.
Check the wear and activity context
Before comparing numbers, record what could change the signal:
| Check | Why it matters | What to log |
|---|---|---|
| Worn finger and hand | Steps are inferred from finger motion | Which hand and finger; changes between days |
| Removal periods | Gaps mean no data, not zero movement | Times off (the study participants logged this too) |
| Activity type | Strength work, cycling and carrying objects move the hand differently from walking | Type and duration of each session |
| Intensity | The lab found larger underestimation as intensity rose | Easy, moderate or hard |
| Comparison device | Wrist, phone and pedometer each measure differently | Where and how it was worn or carried |
Context you log supports hypotheses, not verdicts. A loose fit or a long removal period is a reason to test again, not proof of an error.
A worked audit: why a hub total can exceed both devices
Suppose you wear a ring during the day and carry a phone that also records steps, and both feed a compatible health hub on your phone. This is a generic illustration of interval merging, not a promise of how any specific hub behaves. All numbers are illustrative, not an actual test.
| Source | Interval recorded | Steps |
|---|---|---|
| Ring | 07:00–12:00 and 14:00–22:00 | 6,200 |
| Phone | 12:00–14:00 (ring off for a shower and lunch) | 1,100 |
| Hub total | Merged | 7,300 |
The hub's 7,300 exceeds both 6,200 and 1,100, yet nothing is double counted: it combined non-overlapping intervals. A higher aggregate alone isn't evidence of duplication, and overlapping stored timestamps aren't proof either, because a hub may deduplicate overlapping records when computing its total. To audit:
- Open the hub's steps view and list each source with its time ranges.
- Mark any intervals where two sources both have records.
- If they overlap, check whether the total for that period equals the sum of both sources or only one. Only a total that adds both for the same minutes points to double counting.
- If they don't overlap, the merge is legitimate.
- Compare each device's own total only for the hours it was worn.
If your app offers a setting to choose a preferred step source, consult its help pages and use it when two devices record at once. Calorie totals can also differ simply because one app records only active calories while another adds basal calories. Imports may be delayed if background refresh is off or the apps weren't opened recently. Phone-and-watch reconciliation in specific ecosystems has its own rules and is better handled in separate guides.
Limit cross-device comparisons
Compare like with like:
- Compare steps with steps over the same worn hours, not a day total against a partial day.
- Treat calories and activity time as modeled estimates. Vendors define them differently, so mismatches are expected.
- A logged workout, an imported workout and an automatically detected activity are different records. Check which one you are reading.
- Expect differences to widen at higher intensity, as the laboratory results above suggest.
Why validation results do not transfer across devices
The study above evaluated a smart ring, so its results should not be applied directly to wrist-worn fitness trackers. Different wear positions, motion signals, sensors, and algorithms can affect step counts and activity estimates. For any device, record wear time and activity context, and compare readings only across overlapping time periods using a clearly identified reference. A similar-looking metric does not establish equivalent accuracy across device types.
Take-away checklist
- [ ] I know whether I'm looking at steps, an activity estimate, a logged workout or an imported record.
- [ ] I've noted my device's generation and app version, and I'm not applying another model's error figure to it.
- [ ] I compared only overlapping worn hours.
- [ ] I inspected interval-level sources before calling a higher hub total double counting.
- [ ] I logged removal, activity type and intensity, and treat the log as a hypothesis aid.
- [ ] I read the validation study as strong correlation with systematic differences in one tested setup, not as an accuracy score.
- [ ] I kept recovery and sleep conclusions out of this check.
Accept a result as "consistent" when the difference stays small and stable across several comparable days. If one day is far off, find which interval or activity caused it before blaming the device.
