Accuracy & Validation
How we benchmark every contactless metric against gold-standard reference devices, across skin tones, lighting, and everyday cameras.
How we measure performance you can trust
Circadify estimates vital signs from an ordinary camera, so we hold ourselves to the same yardstick as the equipment clinicians already rely on. Every metric we report is benchmarked against gold-standard medical reference devices captured in sync, and our models are engineered to hold their accuracy across skin tones, lighting, and the everyday cameras people actually use.
Our approach is built on peer-reviewed remote photoplethysmography research and refined with real-world data. Circadify supports monitoring and screening and is not intended for diagnosis.
What our validation is built on
Benchmarks by metric
Each metric below is evaluated against a synchronized reference device. The validated benchmark graphs are being finalized — in the meantime, here's how Circadify performs on each.
Heart Rate vs. Reference Pulse Oximetry
Heart rate is the metric we have the strongest signal for. Circadify recovers the pulse waveform from subtle color changes in the face and converts it to beats per minute, which we compare against a synchronized reference pulse oximeter. In side-by-side recordings the contactless estimate tracks the reference closely as heart rate rises and falls.
Respiratory Rate vs. Reference Monitoring
Respiratory rate is derived from the breathing-induced modulation of the pulse signal together with subtle chest-region motion. We benchmark it against a synchronized reference respiration monitor across both calm and active breathing, where the contactless estimate follows the same cadence.
HRV (SDNN / RMSSD) Agreement
HRV depends on precise beat-to-beat timing. Circadify estimates the inter-beat intervals behind time-domain measures such as SDNN and RMSSD and compares them against ECG-derived reference intervals, capturing the same underlying variability trends.
SpO2 Trend Tracking
SpO2 is estimated from the ratio of pulsatile signals across color channels. We track our trend lines against finger-clip pulse oximetry, where the contactless reading holds its shape and direction as conditions change over a session.
Blood Pressure Trend Tracking
Blood pressure is estimated from pulse-wave timing and waveform morphology. We compare the directional systolic and diastolic trends against cuff-based reference readings captured over the same session to confirm the contactless trend moves with the reference.
Performance Across Skin Tones & Lighting
Robustness matters as much as point accuracy. We evaluate every metric across the full range of skin tones, indoor and outdoor lighting, and different consumer cameras — confirming performance holds in the conditions people actually use, not just an ideal lab setting.
Our validation methodology
We design our evaluations to reflect the messy reality of where vitals are actually measured, so the numbers mean something outside the lab.
Reference-device benchmarking
Every metric is measured against gold-standard medical reference devices captured in sync, giving an honest, side-by-side ground truth.
Diverse demographics
We evaluate across a broad range of skin tones, ages, and body types so performance reflects real, varied populations rather than a narrow sample.
Multiple camera types
Testing spans phone front cameras, laptop webcams, and other consumer sensors to confirm the models generalize beyond a single device.
Real-world lighting
Recordings cover everyday lighting — bright, dim, mixed, and uneven — not only controlled studio conditions, so accuracy holds where people actually are.
Privacy by design
Validation runs on a HIPAA-compliant foundation with end-to-end encryption and BAAs available — raw video never leaves the device.
Continuous model development
Findings feed directly back into model development, so accuracy and robustness keep improving as we gather more real-world data.
Explore further
Get in touch
See how contactless vitals can transform your healthcare delivery.