How every number is computed.
The planned processing chain from raw breath signals to VO₂max and thresholds, written down before validation so you can check it, challenge it and hold us to it.3 Methods v0.1 · planned, not yet validated · last changed 21 Oct 2026.
In development · Pilot planned 2027 · Not for saleDesigned in India
Planned, not frozen.
Design stage. Nothing here has been validated. The algorithm will be frozen and versioned before the validation study, and published in full with the results.
Change log. v0.1 · 21 Oct 2026 · First public draft of the planned methods.
From raw signals to V̇O2.
- SensorsFlow, O₂, CO₂, pressure, temperature, humidity
- SamplingOne device clock for every signal
- Lag correctionFirst-order model per gas sensor
- Breath detectionZero-crossings with hysteresis
- Volume-domain alignmentGas assigned to the air that carried it
- HaldaneInspired volume from the N₂ balance
- BTPS / STPDMeasured pressure, temperature, humidity
- Quality flagsLeak, out-of-range, rejected
- AveragingBreath, rolling or binned; chosen after the test
- Derived variablesV̇O₂, V̇CO₂, RER, V̇E, VT1, VT2, VO₂max
Every step is versioned and will be published. Nothing in this chain is meant to be a black box.
The full Haldane form.
- FIO₂, FICO₂
- Inspired fractions, taken from the room-air reference before each test (dry room air is about 20.93% O₂ and 0.04% CO₂).
- V̇E
- Expired volume per breath, converted to STPD for these equations.
- BTPS
- V × (273 + 37) ÷ (273 + T) × (PB − PH₂O) ÷ (PB − 47 mmHg)
- STPD
- V × 273 ÷ (273 + T) × (PB − PH₂O) ÷ 760 mmHg
- Water vapour
- PH₂O from measured humidity and temperature at the flow sensor, per breath.
What is reported, and how.
| Variable | Reported as | Note |
|---|---|---|
| V̇E, tidal volume | BTPS | Body temperature, ambient pressure, saturated |
| V̇O₂, V̇CO₂ | STPD | 0 °C, 760 mmHg, dry |
| End-tidal O₂ and CO₂ | mmHg | Estimates from the lag-corrected signal |
| Workload | W, or speed and grade | From a trainer or power meter, or entered |
| Heart rate | bpm | From a paired chest strap |
| Time | s | One device clock for breaths, heart rate and workload |
Every breath is stored on its own. Averaging (a number of breaths, a rolling time window, or fixed bins) is applied afterwards and can be changed; VO₂max uses a 30 s rolling average by default. Sampling rates will be set on prototypes and published.
Flagged, never deleted.
- Breath detection
- Breaths are split at flow zero-crossings with hysteresis. Events shorter than a minimum duration or volume (swallows, coughs, partial breaths) are flagged and kept in the export, not deleted.
- Sensor response
- Each gas sensor’s response is modelled as a first-order lag. The signal is corrected with that sensor’s measured time constant before it is used, and the time constant is reported with each test.
- Gas–flow alignment
- Gas concentration is aligned to expired volume, not to clock time, so a breath’s gas is assigned to the air that carried it, whatever the breathing rate.
- V̇O₂ and V̇CO₂
- Classical breath-by-breath calculation from expired volume and expired fractions. Inspired volume is derived with the Haldane (N₂-balance) transformation, which needs measured CO₂ as well as O₂. Inspired flow is measured too and used for breath detection and leak checking.
- Leak check
- Every breath, inspired volume measured by the flow sensor is compared with inspired volume predicted from the N₂ balance. A sustained mismatch flags a probable mask leak, and the flagged breaths are marked in the report and export.
- Corrections
- Ventilation reported at BTPS, gas volumes at STPD, using measured pressure, temperature and humidity.
- End-tidal gases
- End-tidal O₂ and CO₂ are taken from the lag-corrected signal. They are reported as estimates and flagged above the breathing rate at which the sensor can no longer resolve the end-tidal plateau; this limit will be published.
- VT1
- Proposed by the V-slope method and cross-checked against ventilatory equivalents and end-tidal O₂; confirmed or moved by the tester. Both the algorithm’s pick and the final pick are stored.
- VT2
- Detected automatically from the rise in V̇E/V̇CO₂ and fall in PETCO₂; the tester can override, and overrides are logged.
- VO₂max
- Highest 30 s rolling average (window user-selectable). Plateau: an increase of less than 150 ml/min (or less than 50% of the expected rise) over the final 60 s of the ramp. Optional verification bout at 105–110% of peak ramp power after 10 min of recovery. Reported as VO₂peak when neither confirms a maximum.
- Out-of-range flag
- Breaths above the validated flow range are marked in the report and export, and VO₂max is reported with a warning if they fall in the final 60 s.
- Fuel mix
- Only from steady-state stages of 3–5 minutes below VT1 (for example a step test), never from a ramp.
Rejection thresholds (minimum breath duration and volume) will be published with the frozen version.
The N2 balance, every breath.
Nitrogen is neither used nor produced by the body, so the inspired volume can be predicted from the expired volume and the gas fractions. Every breath, that prediction is compared with the inspired volume the flow sensor actually measured. A sustained mismatch means air is escaping around the mask: those breaths are flagged in the report and export, and the flag threshold will be set from prototype data and published.
VT2 automatic. VT1 confirmed by a person.
VT1 is the threshold with the weakest agreement between experts, so the algorithm proposes it (V-slope, cross-checked against ventilatory equivalents and end-tidal O₂) and a tester confirms or moves it. VT2 is detected automatically from the rise in V̇E/V̇CO₂ and the fall in PETCO₂, and can be overridden. The algorithm’s pick, the final pick and who changed it are all stored with the test. VT1 can only be judged after the ramp has passed it, so it is not shown as a live result.
Maximum, or peak.
Highest 30 s rolling average (window selectable). Plateau: an increase of less than 150 ml/min, or less than 50% of the expected rise, over the final 60 s of the ramp. An optional verification bout at 105–110% of peak ramp power after 10 minutes of recovery. Reported as VO₂peak when neither confirms a maximum, and with a warning if out-of-range breaths fall in the final 60 s.
Planned checks, all logged.
Every test: room-air reference for O₂ and CO₂, and a zero-flow check of the flow sensor with the mask held still, alongside the room-air reference. Seal check before each test: planned, once the leak rule is validated. Daily, or when a flow element changes: 3-litre syringe check of flow at several stroke rates (planned; not in the current build). Periodically: two-point gas check with a certified reference gas (about 16% O₂ / 4–5% CO₂; planned, not in the current build). Every check is logged with date, values and pass or fail, and travels with the test in the export. Intervals will be set from prototype drift data and published.
Where it works, and where it doesn’t yet.
| Quantity | Planned | Measured | Validated |
|---|---|---|---|
| V̇E range, current flow element | About 85–100 L/min | — | — |
| V̇E range, high-flow element | Up to about 200 L/min (in development) | — | — |
| Gas response time (10–90%) | To be measured on prototypes | — | — |
| Breathing rate for valid end-tidal values | To be measured on prototypes | — | — |
| Dead space and resistance | To be measured and published | — | — |
| Ambient temperature and humidity | To be defined | — | — |
| V̇O₂ agreement with a cart | Limits registered before testing | — | — |
Known limitations (design stage)
- Flow rangeThe current flow element is characterised to about 85–100 L/min V̇E. Maximal tests in trained athletes need the high-flow element now in development.
- Gas sensor speedOn-mask sensors are slower than cart analyzers. Breath values rely on lag correction, and end-tidal values are estimates whose valid breathing-rate range will be published.
- VT1 needs a personVT1 is proposed by the algorithm and confirmed by a tester. VT2 is automatic but can be overridden.
- Mask sealLeaks bias every gas-exchange variable. Leak-flagged breaths are marked, not hidden.
- Dead space and resistanceThe mask and flow element add both. Figures will be published; their effect is largest at rest and at very high ventilation.
- EnvironmentExtreme cold, heat, humidity, rain and altitude are untested.
- ProtocolsFuel mix and energy expenditure need steady-state stages; thresholds from step tests are less precise than from a ramp.
- AccuracyUnknown until validation is published.
Registered before the first test.
The validation plan sets the reference cart, the bench tests for each flow element, the acceptance limits, the statistics and the publication rules. It will be registered publicly before the first participant is tested, and linked here once the registry entry exists.
Two levels, both yours.
Breath table: every breath with time, volumes (BTPS), fractions, end-tidal values, V̇O₂ and V̇CO₂ (STPD), quality flags (leak, out-of-range, rejected), heart rate and workload. Raw signals: flow, O₂, CO₂, pressure, temperature and humidity at the native sample rate, uncorrected and corrected, so you can rerun every step yourself. Each file carries its metadata: device serial, firmware and algorithm version, flow element, check log, ambient conditions, mask size and protocol. Format: CSV with a JSON metadata sidecar; schema and units documented and versioned.
Example breath table (CSV, example data) Example breath table (Excel, formatted) Example metadata sidecar (JSON)
Planned, in writing.
Open methods.
Every processing step, from breath detection to threshold picking, published as documented equations and a reference implementation, with a version number on every report. When an algorithm changes, the change log says what changed and how it affects results, and old data can be reprocessed with any version.
Open validation data.
De-identified breath-by-breath and raw-signal data from the validation study, published with the paper, so anyone can check our numbers.
What these methods build on.
- Beaver WL, Wasserman K, Whipp BJ. A new method for detecting anaerobic threshold by gas exchange. J Appl Physiol. 1986;60(6):2020–2027.
- Wasserman K, Hansen JE, Sue DY, et al. Principles of Exercise Testing and Interpretation. Lippincott Williams & Wilkins.
- Poole DC, Jones AM. Measurement of the maximum oxygen uptake V̇O₂max: V̇O₂peak is no longer acceptable. J Appl Physiol. 2017;122(4):997–1002.
- American Thoracic Society; American College of Chest Physicians. ATS/ACCP statement on cardiopulmonary exercise testing. Am J Respir Crit Care Med. 2003;167(2):211–277.
- Hopkins WG. Measures of reliability in sports medicine and science. Sports Med. 2000;30(1):1–15.
- Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1(8476):307–310.
- Atkinson G, Nevill AM. Statistical methods for assessing measurement error (reliability) in variables relevant to sports medicine. Sports Med. 1998;26(4):217–238.
- Van Hooren B, Souren T, Bongers BC. Accuracy of respiratory gas variables, substrate, and energy use from 15 CPET systems during simulated and human exercise. Scand J Med Sci Sports, 2024.
Disagree with a choice here? Tell us before we freeze it.
Named reviewers will be acknowledged in the methods paper. Email the founder with the section and your reasoning.