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.

28,516
Total sessions
815
Individual athletes
5
Datasets
4-6 bpm
Median MAE
3
Parameters

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

Download PDF ↓

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

Download PDF ↓

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

Download PDF ↓

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 TrueZoneApple (Nazaret et al. 2023)
Fitted MAE7.0 bpm7.22 bpm
MAPE4.3%4.2%
Parameters3 (interpretable)32-dim latent embedding
Training dataNone (per-user Bayesian)270,000 runs / 7,465 users
Model typePhysiological ODEHybrid ODE + causal CNN
Derived metricsFull physiological profileHeart rate prediction only
Endurance modellingYes (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.

DatasetSubjectsSessionsAccuracySource
Running (Endomondo)3477,2934.0 bpm median MAEUCSD FitRec, CC-BY-NC
Running (Apple benchmark)752,9706.9 bpm median MAEEndomondo subset, Nazaret comparison
Cycling (GoldenCheetah)29713,8876.0 bpm median MAEGoldenCheetah OpenData, CC0
Football (Elite Women)136846-month season trackedSoccerMon dataset, Zenodo
ICE-TRACK (Mixed Sport)833,682≤5% MAPEIcelandic recreational athletes
Total81528,5164–6 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.