Science · Validation · Preliminary results

Sensor Bio V2 validation

Preliminary results for algorithm library version 8.11.0: how closely Sensor Bio V2 readings match a reference device worn at the same time, metric by metric.

Algorithm library version 8.11.0Preliminary resultsOctober 2026

How to read these results

Preliminary results for algorithm library version 8.11.0, conducted by an independent university exercise physiology laboratory. Each Sensor Bio V2 metric was compared against a reference worn at the same time, by healthy adults aged 20 to 60, with a device on each wrist. Steps were tested on video-counted walking and whole days of normal wear.

Download the preliminary one-page summary (PDF) 202 KB

Average error
How far a reading is from the reference, on average. Smaller is closer.
Coverage
How often the device gave a reading during the recording.
Within a tolerance
The share of readings that landed within a set distance of the reference, such as within 5 bpm.
r
Correlation with the reference. A value of 1 means the two move together exactly.
01Rest and exercise

Heart rate

At rest, Sensor Bio V2 heart rate was on average 1.58 bpm from a clinical 12-lead ECG, with 94.9% of seconds within 5 bpm. In a hard exercise session the average error was 4.08 bpm, with 87.4% of readings within 10%. Steady efforts are the most accurate: 1.8 bpm average error while cycling.

How often A reading every second, with 99.85% coverage at rest and 99.19% in exercise.

1.58bpm

Average error at rest. 94.9% of seconds within 5 bpm.

4.08bpm

Average error in a hard exercise session. 87.4% within 10%.

Agreement with the reference

  • 94.9% of seconds within 5 bpm at rest

  • 87.4% within 10% in hard exercise

Coverage

  • 99.85% coverage at rest

  • 99.19% coverage in exercise

Reference
Clinical 12-lead ECG
Participants and recordings
32 participants, 118 recordings
Conditions
Rest, and a hard exercise session. Cycling reported separately.

Keep in mind

  • The exercise session changes intensity almost every minute, which makes it a hard case. Steady activity is more accurate.
02Rest

Heart rate variability (HRV)

Sensor Bio V2 reports HRV as RMSSD over 45 seconds, about every 10 seconds. At seated rest, each reading was on average 6.73 ms from a clinical 12-lead ECG, with 82.8% within 10 ms. Averaged over a session, the error was 3.9 ms, with a correlation of r = 0.98.

How often A 45-second RMSSD about every 10 seconds, with 93.9% coverage.

6.73ms

Average error per 45-second RMSSD reading. 82.8% within 10 ms.

3.9ms

Average error of the session-average HRV (r = 0.98).

Agreement with the reference

  • 82.8% of readings within 10 ms

Coverage

  • 93.9% coverage

Reference
Clinical 12-lead ECG
Participants and recordings
30 participants, 60 recordings
Conditions
Seated rest

Keep in mind

  • Not reliable with frequent ectopic beats.
  • Validated at seated rest only.
03Rest

Beat-to-beat intervals

Sensor Bio V2 reports every beat-to-beat interval. At seated rest, intervals were on average 5.17 ms from a clinical 12-lead ECG, with 89.2% within 10 ms and a correlation of r = 0.999. The device detected 86% of ECG heartbeats within 60 ms, with 4.5% false beats. When it is wrong, it almost always leaves a beat out rather than adding one.

How often Every beat-to-beat interval is reported.

5.17ms

Average interval error. 89.2% within 10 ms (r = 0.999).

86%

Of ECG heartbeats detected within 60 ms. 4.5% false beats.

Agreement with the reference

  • 89.2% of intervals within 10 ms

  • 86% of ECG heartbeats detected within 60 ms

Reference
Clinical 12-lead ECG
Participants and recordings
26 participants, 48 recordings
Conditions
Seated rest

Keep in mind

  • 4.5% false beats. Misses are far more common than added beats.
  • Validated at seated rest only.
04Rest

Breathing rate

Sensor Bio V2 derives breathing rate from the heartbeat rhythm, once a second. At seated rest, each reading was on average 1.15 breaths per minute from a Cosmed metabolic cart with a face mask, with 92.4% within 3 breaths per minute. Averaged over a session, the error was 0.77 breaths per minute.

How often Once a second, with 90.3% coverage. Reports 6 to 40 breaths/min.

1.15br/min

Average error per reading. 92.4% within 3 breaths/min.

0.77br/min

Average error of the session-average breathing rate.

Agreement with the reference

  • 92.4% of readings within 3 breaths/min

Coverage

  • 90.3% coverage

Reference
Cosmed metabolic cart (face mask)
Participants and recordings
30 participants, 60 recordings
Conditions
Seated rest

Keep in mind

  • Reports 6 to 40 breaths/min.
  • Validated at seated rest only.
05Rest and exercise

Skin temperature

Sensor Bio V2 skin temperature was on average 0.58 °C from a skin-temperature patch on the same wrist at rest, and 0.59 °C in exercise. 84% of readings were within 1 °C of the reference, at rest and in exercise.

How often A reading every minute, with 99% coverage.

0.58°C

Average error at rest (0.59 °C in exercise).

84%

Of readings within 1 °C of the reference, at rest and in exercise.

Agreement with the reference

  • 84% of readings within 1 °C

Coverage

  • 99% coverage

Reference
Skin-temperature patch on the same wrist
Participants and recordings
31 participants, 111 recordings
Conditions
Rest and exercise

Keep in mind

  • Allow about 15 minutes after putting the device on for the sensor to settle.
06Everyday walking

Steps

On video-counted walking (OxWalk), the Sensor Bio V2 step algorithm counted 93.1% of steps, with an F1 score of 84.4%. Over whole days of normal wear, the average day-total difference was +1.7% compared with other wearables, and 21 of 33 days were within ±10%.

How often A step total every minute.

93.1%

Of video-counted steps counted (OxWalk). F1 84.4%.

+1.7%

Average day-total difference vs. other wearables. 21 of 33 days within ±10%.

Agreement with the reference

  • 93.1% of video-counted steps counted

  • 21 of 33 days within ±10% of other wearables

Reference
Steps counted by hand from video, and other wearables
Participants and recordings
39 video recordings, and 33 days of normal wear
Conditions
Video-counted walking, and whole days of normal wear

Keep in mind

  • Video accuracy uses the device's step algorithm on research wrist sensors.
  • The day comparisons were recorded on a Sensor Bio device.

Good to know

  • Wear

    Fasten snugly, one to two finger widths above the wrist bone. Either wrist performs the same.

  • Hard-case exercise

    The heart rate exercise session changes intensity almost every minute. Steady activity is more accurate.

  • Rest-only metrics

    HRV, beat-to-beat intervals and breathing rate were validated at seated rest.

  • Population

    Healthy adults at one site, including a range of skin tones. Children and people with cardiac or lung conditions were not studied.

Methods

How we test

  1. 01

    Side by side with a reference

    Each metric was compared against a reference worn at the same time.

  2. 02

    University laboratory studies

    Healthy adults aged 20 to 60 took part at one site, including a range of skin tones, with a device on each wrist.

  3. 03

    Matched to the metric

    A clinical 12-lead ECG for heart rate, HRV and beat-to-beat intervals. A Cosmed metabolic cart with a face mask for breathing rate. A skin-temperature patch on the same wrist for skin temperature. Hand-counted video and other wearables for steps.

  4. 04

    Real conditions

    Heart rate and skin temperature were tested at rest and in exercise. HRV, beat-to-beat intervals and breathing rate at seated rest. Steps on video-counted walking and whole days of normal wear.

  5. 05

    Tied to an algorithm library version

    These are preliminary results for algorithm library version 8.11.0 on Sensor Bio devices. Results are listed by algorithm library version.

References by metric

  • Heart rate

    Clinical 12-lead ECG

    32 participants, 118 recordings. Rest, and a hard exercise session. Cycling reported separately..

  • HRV

    Clinical 12-lead ECG

    30 participants, 60 recordings. Seated rest.

  • Beat-to-beat

    Clinical 12-lead ECG

    26 participants, 48 recordings. Seated rest.

  • Breathing rate

    Cosmed metabolic cart (face mask)

    30 participants, 60 recordings. Seated rest.

  • Skin temperature

    Skin-temperature patch on the same wrist

    31 participants, 111 recordings. Rest and exercise.

  • Steps

    Steps counted by hand from video, and other wearables

    39 video recordings, and 33 days of normal wear. Video-counted walking, and whole days of normal wear.

Questions

FAQ

Are these final results?

No. These are preliminary results for algorithm library version 8.11.0, measured under the conditions tested by an independent university exercise physiology laboratory. The figures may change as the algorithm develops.

How accurate is Sensor Bio V2 heart rate?

On algorithm library version 8.11.0, Sensor Bio V2 heart rate was on average 1.58 bpm from a clinical 12-lead ECG at rest, with 94.9% of seconds within 5 bpm. In a hard exercise session that changes intensity almost every minute, the average error was 4.08 bpm. Steady efforts are the most accurate: 1.8 bpm average error while cycling.

How accurate is Sensor Bio V2 HRV?

At seated rest, each 45-second RMSSD reading was on average 6.73 ms from a clinical 12-lead ECG, with 82.8% within 10 ms. The session-average HRV was within 3.9 ms on average (r = 0.98). HRV is not reliable with frequent ectopic beats.

What do average error and coverage mean?

Average error is how far a reading is from the reference. Coverage is how often the device gave a reading.

What reference devices were used?

A clinical 12-lead ECG for heart rate, HRV and beat-to-beat intervals, a Cosmed metabolic cart with a face mask for breathing rate, a skin-temperature patch on the same wrist for skin temperature, and hand-counted video and other wearables for steps.

Who took part in the validation studies?

Healthy adults aged 20 to 60 in university laboratory studies at one site, including a range of skin tones. Children and people with cardiac or lung conditions were not studied.

Were HRV and breathing rate validated during exercise?

No. HRV, beat-to-beat intervals and breathing rate were validated at seated rest. Heart rate and skin temperature were tested at rest and in exercise.

How should I wear Sensor Bio V2 for the best readings?

Fasten it snugly, one to two finger widths above the wrist bone. Either wrist performs the same. For skin temperature, allow about 15 minutes after putting the device on for the sensor to settle.

Is Sensor Bio V2 a medical device?

No. Sensor Bio V2 is a general wellness device. It is not FDA cleared and is not intended to diagnose, treat, cure, or prevent any disease.

Sensor Bio V2 is a general wellness device. It is not FDA cleared and is not intended to diagnose, treat, cure, or prevent any disease. These are preliminary results for algorithm library version 8.11.0, under the conditions tested. They show how closely readings agree with a reference and are not a measure of health.

Validation history

  • Algorithm library version 8.11.0, October 2026 (current) · Preliminary results · PDF

Sensor Bio V2

Build on signal you can check

Heart rate, HRV, beat-to-beat intervals, breathing rate, skin temperature and steps, from one wrist band, with raw PPG for teams that need it.