Validation
Lab-validated. Benchmark-tested. Scaled to 28,000+ sessions.
TrueZone has been validated against gold-standard cardiopulmonary exercise testing, benchmarked head-to-head against Apple's neural network, and fitted to five independent real-world datasets across running, cycling, and team sports.
Evidence standard
Same model architecture across lab tests, public datasets, and field sport.
No sport-specific tuning. No dataset-specific parameter space. Just the same physiological model tested from controlled labs to ordinary field data.
Laboratory Validation
Tested against gold-standard CPET.
TrueZone's outputs have been compared against cardiopulmonary exercise testing (CPET) and metabolic biomarkers in controlled laboratory studies at the University of Iceland and Reykjavík University.
METFIT PhD Study - University of Iceland - 2025
Validation of a Heart Rate-Based System for Assessing Exercise Thresholds and Aerobic Endurance
38 recreational runners completed submaximal treadmill tests, 50-metre sprints, and incremental CPET. TrueZone estimates showed strong agreement with CPET and no significant difference for VO2max, Vmax, or VT2 speed.
Steinarsson A, Jakobsdottir G, Johannsson E.
VO2max
R2 = 0.86
HRmax
R2 = 0.86
VT2 speed
R2 = 0.91
METFIT PhD Study - University of Iceland - 2025
Metabolic Tracking Using Heart Rate: A Model for Energy Balance and Glycemic Control
8 healthy adults were tested for fasting metabolic biomarkers, controlled exercise trials, and a 7-hour postprandial feeding trial. Energy expenditure, intake, and MFI were compared against reference measures.
Steinarsson A, Jakobsdottir G, Fridriksson JH, Johannsson E.
EE accuracy
5.8% MAE
EI accuracy
4.8% MAE
MFI vs GRI
R2 = 0.99
Reykjavik University - 2019-2024
Four independent validation studies
A series of Reykjavik University studies validated Driftline's heart-rate kinetics model across different populations and testing contexts, comparing TrueZone outputs against laboratory measurements.
Steinarsson A. Summary report.
IAK
2019
Masters
2021
Elderly Fitness
2024
Benchmark
Driftline vs. Apple: same accuracy, radically different approach.
Nazaret et al. (2023) trained a hybrid ODE + deep neural network on 270,000 runs from 7,465 Apple Watch users. TrueZone matches their heart rate prediction accuracy with a pure physiological model and no global training data.
| Driftline TrueZone | Apple (Nazaret et al. 2023) | |
|---|---|---|
| Fitted MAE | 7.0 bpm | 7.22 bpm |
| MAPE | 4.3% | 4.2% |
| Parameters | 3 (interpretable) | 32-dim latent embedding |
| Training data | None (per-user Bayesian) | 270,000 runs / 7,465 users |
| Model type | Physiological ODE | Hybrid ODE + causal CNN |
| Derived metrics | Full physiological profile | Heart rate prediction only |
| Endurance modelling | Yes (E, 0–100%) | No |
Real-World Scale
Five independent datasets. Three sports. One model.
The same three-parameter architecture fits all five datasets without sport-specific tuning. Parameters converge within 3–10 sessions from ordinary activity data.
| Dataset | Subjects | Sessions | Accuracy | Source |
|---|---|---|---|---|
| Running (Endomondo) | 347 | 7,293 | 4.0 bpm median MAE | UCSD FitRec, CC-BY-NC |
| Running (Apple benchmark) | 75 | 2,970 | 6.9 bpm median MAE | Endomondo subset, Nazaret comparison |
| Cycling (GoldenCheetah) | 297 | 13,887 | 6.0 bpm median MAE | GoldenCheetah OpenData, CC0 |
| Football (Elite Women) | 13 | 684 | 6-month season tracked | SoccerMon dataset, Zenodo |
| ICE-TRACK (Mixed Sport) | 83 | 3,682 | ≤5% MAPE | Icelandic recreational athletes |
| Total | 815 | 28,516 | 4–6 bpm median |
All datasets use the same model architecture and parameter space. No sport-specific tuning, no dataset-specific adjustments. The model fits running with GPS speed, cycling with power meter data, and team sports with GPS vest tracking—demonstrating true universality across input modalities and movement patterns.
Three parameters. Lab-validated. 28,000+ sessions.
From gold-standard CPET comparison to large-scale real-world datasets, TrueZone's accuracy is transparent and reproducible.