Honest about accuracy
We publish full numbers — including where we are not yet where we want to be. Field performance for camera-based heart rate is typically lower than curated lab sets. We will not quote a single headline percentage as if uncertainty did not exist.
What the product shows today
Every reading in the cabinet carries a signal quality score. When confidence is too low, the metric is withheld with an explanation — not replaced by a guess.
How each vital is measured — camera, microphone, estimate, or lab log — is on the technology page.
By vital
One row per vital, filled from published model_versions rows (beta or published only). N is paired ground-truth samples. MAE is taken from a stored validation report — never invented. Until that report exists, the cells stay blank.
| Vital | N | MAE | Status |
|---|---|---|---|
| Heart rate (camera PPG)Quality shown on every reading. Field MAE is not published yet. | — | — | Collecting first paired data |
| HRV (camera PPG)Withheld when quality is too low. No field MAE yet. | — | — | Collecting first paired data |
| Breathing rate (camera)Optical estimate from the face scan. No field MAE yet. | — | — | Collecting first paired data |
| Stress load (composite)No reference device. Compared with your usual, not a named condition. | — | — | Wellness estimate |
| Oxygen (camera estimate)Marked as an estimate. Paired against a pulse oximeter when you log one. | — | — | Collecting first paired data |
| Blood pressure (camera estimate)Prefer a cuff or watch. Camera figure is an estimate until paired data exists. | — | — | Collecting first paired data |
| Hemoglobin (camera estimate)Lab result is the precise figure. Camera level is an estimate. | — | — | Collecting first paired data |
| HbA1c (lab / manual)No camera figure. Trend from logs and lab entries only. | — | — | Lab / manual |
N and MAE stay blank until a beta or published row exists in model_versions. Training and shadow heads never appear here. Empty is unpublished, not zero. We do not invent error statistics.
Stratified validation — forthcoming
When validation completes, this page will carry error bands, coverage, and breakdowns by skin tone, motion, and heart-rate range — with methodology and limitations spelled out. The structure below is fixed now; the numbers are not invented.
| Metric · stratum | MAE | RMSE | r | Coverage | Validated |
|---|---|---|---|---|---|
| Heart rate (camera PPG)All conditions | Pending | Pending | Pending | Pending | — |
| Heart rate (camera PPG)Fitzpatrick I–III | Pending | Pending | Pending | Pending | — |
| Heart rate (camera PPG)Fitzpatrick IV–VI | Pending | Pending | Pending | Pending | — |
| Heart rate (camera PPG)Still / seated / hand motion | Pending | Pending | Pending | Pending | — |
| Heart rate (camera PPG)HR <60 / 60–100 / >100 bpm | Pending | Pending | Pending | Pending | — |
Coverage = share of attempts that yielded a result. We publish it beside error — a product that always returns a number never tells you when it does not know.
Limitations we already know
- ·Camera PPG degrades with motion, poor lighting, and darker skin tones — we stratify for this honestly.
- ·Lab-curated datasets overstate field performance; we will publish field numbers, not textbook sets.
- ·Heart-sound validation is separate and ships when the cardio channel is validated on device.
We do not claim AAMI EC13 equivalence or compare ourselves to cleared medical devices — that would be a clinical claim this wellness product does not make.
© 2026 EDMC. All rights reserved.