Start with the label, not the leaderboard
Imagine two unnamed apps showing different HRV values. This is an invented report-reading example, not a device test or a person’s health record. Before deciding that one app is wrong, find the full name of each measure. A shared dashboard heading does not establish that the values answer the same question.
HRV describes variability in intervals between heartbeats. RMSSD summarizes successive interval differences; SDNN describes the spread of normal-to-normal intervals. Both can be reported in milliseconds while using different calculations. Stone and colleagues describe these distinctions in their original device-validation paper. A unit tells you the scale, not the complete definition.
What exactly was calculated, over which recording, and under what conditions?
Many trials do not mean many people
Stone and colleagues compared seven commercial device/application configurations with multi-lead ECG during seated rest. Their 148 trials came from five healthy young adults, not 148 different people. Reported RMSSD accuracy varied across configurations. Recordings lasted three or five minutes depending on the configuration.
The researchers matched recording periods using manual timestamps and examined processed outputs. Retained trial counts differed after outlier removal. This small historical study illustrates why the tested setup belongs beside an accuracy claim; it is not a ranking of today’s products or proof that every optical measurement is unsuitable.
Good agreement still has a setting
Coste and colleagues simultaneously recorded an ECG-based H10 chest strap and an optical OH1 sensor in 31 healthy adults aged 21–66. Seated and lying-down conditions were tested in randomized order. They found generally good agreement, with wider inter-device limits of agreement when seated. Their two-minute segments were extracted from the five-minute recordings, rather than collected as independent sessions.
Only people with light skin phototypes I–III were included, and each posture lacked repeated recordings. The authors therefore could not fully separate posture from natural fluctuation. These findings do not validate every optical device, overnight summary, exercise recording, skin tone or clinical use.
Agreement describes the tested measurement and conditions. It is not a universal device certificate.
Three checks before comparing two values
First, match the metric and its units. Look for the actual calculation name rather than a generic HRV or recovery label. If one report omits it, record the definition as unknown. Do not assume that two nearby-looking values share a method.
Second, match the window. Ask which part of a recording contributed to the value, its duration, and whether it represents a selected segment or a summary. A display date alone cannot answer those questions.
Third, match the context and processing. Ask about the sensor, recording conditions and handling of unusable intervals. Missing details are a reason to limit a comparison, not evidence that a particular device made an error. These three checks are our research-reading framework, not a clinically validated assessment.
Preserve the question instead of forcing an answer
A useful note beside an export can be simple: metric identified; window known or unknown; collection and processing context documented or missing. Keep those notes with the values when discussing a report or asking its provider for an explanation. This is a documentation task, not a request to perform a new physiological test.
The studies here do not provide a personal correction factor, a way to average incompatible values, or a threshold that diagnoses stress or recovery. They also cannot determine why your own number changed. A more useful takeaway is a sharper question: are these two reports describing the same measurement?