Research · validation roadmap (pre-publication)
The pain:Most wearables make claims they can’t prove. We publish the validation plan before the bracelet ships.
Six validation studies are in design, with target cohorts and target accuracy brackets pre-registered. The bracelet ships first; the studies run on the live cohort; methods publish on peer-review acceptance. This page lists the plan today, the results when they exist.
0
Studies in design
0+
Target cohort size
0+
User-night plan
r = 0.00
HRV target vs clinical ECG
§ 01 · Pre-registered targets
These are not findings. They are the acceptance brackets we publish in advance — once each study completes peer review, the published numbers replace the targets here, with the original target left visible for comparison.
target r ≥ 0.00
HRV vs ECG
Concordance, RMSSD
target κ ≥ 0.00
Sleep staging
Cohen's kappa, 4-class
target AUC ≥ 0.00
Recovery prediction
Next-day, overnight inputs
target ≥ 0%
Cortisol reduction
90-day cohort, opt-in
§ 02 · The six designed studies
Each study validates one part of the Wellex stack: sensors, sleep, longitudinal outcomes, hardware architecture, recovery modelling, WVI personalisation. Targets and protocols are fixed in advance; results replace targets at peer-review acceptance.
Designed protocol: double-blind validation, target n ≥ 300 participants aged 18–72, 14 consecutive days, during rest, exercise and sleep. Bland-Altman agreement analysis and Lin's concordance correlation coefficient against reference 12-lead ECG.
Designed protocol: Wake / Light / Deep / REM classification from a multi-sensor fusion model (optical waveform + 3-axis motion + skin temperature + SpO₂ variability). Target training set ≥ 10,000 polysomnography-labelled nights, hold-out ≥ 2,000 nights.
Designed protocol: prospective 90-day study, target n ≥ 1,000 participants. The pre-registered hypothesis tests whether sustaining a 10-point WVI improvement correlates with measurable cortisol reduction and self-reported energy gains. Cohort recruits at pool launch.
Engineering paper on the 8-sensor bracelet architecture. Eliminating the display frees roughly 40% more internal volume for battery and allows continuous 10 Hz sampling across every channel — delivering target 14-day battery life under continuous operation.
Designed protocol: gradient-boosted ensemble trained on the active-cohort dataset. Pre-registered hypothesis: overnight HRV trend, respiratory rate, skin temperature and sleep-stage distribution predict below-baseline recovery mornings with target sensitivity ≥ 0.80 and specificity ≥ 0.75.
Methodology paper on the 14-day adaptive baseline. Personalised WVI thresholds account for age, sex, resting HRV, habitual sleep and activity level. Bias-reduction targets pre-registered: scoring bias across age and BMI subgroups must reduce ≥ 50% vs population-fixed thresholds.
§ 03 · Sensor accuracy vs the category
The gap between consumer wearables and clinical reference equipment has been narrowing for years. For the core vital signs, Wellex is where it closes.
Signal | Clinical reference | Wellex | Typical consumer wearable | |
|---|---|---|---|---|
| row-0 | HRV (RMSSD) | 12-lead ECG | target r ≥ 0.90 | r ≈ 0.80 · ±8 ms |
| row-1 | Sleep staging | Polysomnography | target κ ≥ 0.75 · 4-class | κ ≈ 0.55 · 2-class |
| row-2 | Next-day recovery | Self-report + HRV panel | target AUC ≥ 0.85 | Rarely reported |
| row-3 | SpO₂ | Pulse oximetry | target ±2 % MAE | ±2.5–4 % MAE |
| row-4 | Skin temperature trend | Medical thermistor | target ±0.15 °C | ±0.3 °C |
| row-5 | Continuous 10 Hz sampling | Research-grade rig | 14-day battery target | 1–2 day battery |
§ 04 · How we validate
Every sensor claim and every WVI correlation passes through clinical-grade reference equipment in controlled studies before publication or release.
No feature ships on internal benchmarks alone. Every accuracy figure on this page is validated against clinical reference equipment in controlled studies before publication.
Full algorithmic methodology is published: training data composition, model architecture, known limitations. Open methodology earns more trust than marketing claims.
WVI is not a static algorithm. We run 90-day cohort studies on a rolling basis to verify that score changes correlate with real health outcomes. Calibration updates publish quarterly.
Research direction is set by an external clinical advisory board: cardiology, sleep medicine, sports science, biostatistics. Board members review all materials before release.
Want to see the engine, not just the evidence?
All studies are conducted in accordance with the Declaration of Helsinki. Participant data is anonymised and processed under Wellex's GDPR-compliant privacy framework. For research collaboration inquiries, write to [email protected].