> ## Documentation Index > Fetch the complete documentation index at: https://developer.neurofit.app/llms.txt > Use this file to discover all available pages before exploring further. # Metrics > What each metric means, its units, the quality tier, and the validated accuracy against a Polar H10 chest strap. One 60-second reading returns the metrics below in a `VitalsResult`. Readings are processed entirely on the phone with deterministic signal processing, not machine learning. No images are stored or transmitted; the SDK returns only these metrics, a quality tier and diagnostics. The NEUROFIT Vitals SDK is not a medical device. The metrics describe autonomic and cardiorespiratory state for wellness and self-tracking; they do not diagnose, treat or prevent any condition. ## Result fields | Field | Unit | Meaning | | - | - | - | | `heartRateBpm` | beats per minute | Mean heart rate over the reading. | | `rmssdMs` | ms | HRV (RMSSD): the root mean square of successive differences between normal heartbeat intervals. The HRV number wearables report. Higher generally reflects more parasympathetic (rest-and-digest) activity. | | `sdnnMs` | ms | SDNN: the standard deviation of normal heartbeat intervals over the reading. Reflects overall variability. | | `baevskyStressIndex` | dimensionless | Baevsky stress index. Higher values reflect a more rigid, sympathetic-dominant rhythm. Often shown on a log scale. | | `sd2sd1` | ratio | Poincaré SD2:SD1. The ratio of long-term to short-term variability. It is derived from SDNN and RMSSD (SD1 = RMSSD / sqrt 2; SD2 = sqrt(2 SDNN squared minus SD1 squared)) and is absent when either input is missing or the expression is undefined. | | `breathingRateBrpm` | breaths per minute | Breathing rate over the reading. Returned at any confidence. | | `breathingRateConfidence` | 0 to 1 | Confidence in the breathing rate. Present exactly when `breathingRateBrpm` is present (0.0 when the engine gives none). Below 0.5 the rate is still returned; label it an estimate or hide it, your choice. | | `confidence` | 0 to 1 | Confidence in the primary metrics: the HRV confidence, falling back to the heart-rate confidence, else 0. | | `quality` | tier | `clean` (3), `usable` (2) or `withheld` (1). See below. | | `flags` | list of strings | Reasons the engine attached to the reading: `hrv_unreliable_beats`, `too_short_for_freq_hrv`, `too_short_for_rr`, `insufficient_contact`, `low_quality`, `consistency_gate_rejected`. | | `diagnostics` | object | Capture and quality internals for support and analytics. See [Diagnostics](/diagnostics). | | `id`, `capturedAt`, `durationSeconds` | | A UUID (iOS `UUID`, Android `String`), the completion time (iOS `Date`, Android epoch milliseconds; ISO 8601 in the wrappers) and the reading duration in seconds. | | `engineVersion`, `sdkVersion` | | `1.3.0` and `1.0.0` for this release. Store them with the result. | Metrics are optional (`nil` / `null`) when the engine could not recover them; every value is finite (no NaN reaches a result or its JSON). Heart rate is the one metric a completed reading always has; a reading with no heart rate ends in `failed(noHeartRate)` rather than a result. During the reading, `LivePreview` carries provisional values: heart rate from the first seconds, HRV (RMSSD) from about 20 s, breathing rate from about 30 s. They are for display only; store the final `VitalsResult`. ## Quality tier The SDK grades every reading and withholds the ones it cannot stand behind. | Tier | Meaning | What to do | | - | - | - | | `clean` | High-quality signal throughout. | Show the metrics. | | `usable` | Good signal with some imperfect stretches. Accuracy figures below include both `clean` and `usable` readings. | Show the metrics. | | `withheld` | The SDK is not confident in the HRV metrics. `rmssdMs`, `sdnnMs`, `baevskyStressIndex` and `sd2sd1` may be absent. | Do not show HRV numbers. Offer a retry. Do not store the reading as an outcome. | Two rules apply on top of the engine's grading. A reading whose values are outside physiological plausibility (heart rate outside 20 to 200 bpm, HRV (RMSSD) outside 5 to 400 ms, breathing rate above 60 breaths per minute; an absent breathing rate counts as 0) is downgraded to `withheld` and `sd2sd1` and `baevskyStressIndex` are dropped. And a breathing rate with confidence below 0.5 is still returned with its confidence, so the host decides how to present it. In the validation study the SDK withheld 3.3% of readings. ## Validated accuracy The figures come from the NEUROFIT validation study: 30 participants, 265 accepted readings, each a 60-second seated reading taken while wearing a Polar H10 chest strap, with the strap's beat-to-beat timing as the reference. Half the readings were at rest and half during paced breathing at 6 breaths per minute. Three readings with frequent irregular heartbeats were excluded. The two investigators are excluded from every figure. | Metric | Mean absolute error | r | 95% limits of agreement | Notes | | - | - | - | - | - | | Heart rate | 0.46 bpm | 0.999 | -1.4 to +0.9 bpm | 90% of readings within 1 bpm | | HRV (RMSSD) | 3.39 ms | 0.979 | -9.1 to +9.8 ms | 94.7% within 10 ms, none off by 20 ms or more | | SDNN | 4.13 ms | 0.984 | -9.8 to +12.0 ms | | | Baevsky stress index (ln) | 0.18 | 0.970 | -0.51 to +0.42 | 76% of readings within 25% of the strap value | | Poincaré SD2:SD1 | 0.28 | 0.908 | -0.81 to +0.77 | Error tracks RMSSD error | | Breathing rate (paced 6 brpm) | 0.46 brpm | n/a | n/a | 89% within 1 brpm, confidence-gated | ### Consistent across groups No group had a reading off by 20 ms or more. Groups with one or two people describe those individuals, not the whole group. | Group | Participants | Readings | RMSSD error (ms) | Within 10 ms | | - | - | - | - | - | | iOS | 21 | 178 | 2.81 | 97% | | Android | 9 | 87 | 4.58 | 91% | | Resting | 30 | 134 | 2.98 | 96% | | Paced breathing | 30 | 131 | 3.80 | 93% | | Monk skin tone 2 | 4 | 33 | 2.16 | 97% | | Monk skin tone 3 | 3 | 28 | 1.48 | 100% | | Monk skin tone 4 | 5 | 46 | 4.14 | 91% | | Monk skin tone 5 | 9 | 79 | 2.97 | 96% | | Monk skin tone 6 | 6 | 53 | 5.13 | 91% | | Monk skin tone 8 | 3 | 26 | 3.39 | 96% | | Female | 12 | 105 | 3.50 | 95% | | Male | 18 | 160 | 3.32 | 94% | | Age 18–24 | 1 | 8 | 2.00 | 100% | | Age 25–34 | 21 | 181 | 3.66 | 94% | | Age 35–44 | 5 | 47 | 3.37 | 91% | | Age 45–54 | 1 | 10 | 1.01 | 100% | | Age 55–64 | 2 | 19 | 2.70 | 100% | The full method and plots are in the [validation report (PDF)](https://neurofit.app/assets/pdf/neurofit-vitals-sdk-validation-report-2026-09.pdf). These are interim development results; the finished system is being tested blind on new participants, and the figures will be updated as the study grows. ## Limits of the validation The accuracy figures apply to **seated, still, 60-second readings**. Not yet claimed: * Accuracy for Monk skin tones 7, 9 and 10 (no participants in the study so far). * Accuracy during movement. * Breathing rate outside paced breathing at 6 breaths per minute. * Readings shorter or longer than 60 seconds. Most participants were 25 to 34 years old. ## Presenting the numbers * Round heart rate and HRV (RMSSD) to whole numbers, breathing rate to one decimal. * Say "HRV (RMSSD)" the first time you show HRV, so users comparing with a wearable know which HRV it is. * Compare a person with their own baseline, not with population tables; see [Outcomes best practices](/outcomes-best-practices). * When `quality` is `withheld`, show a calm retry prompt rather than a number. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.