Metric VBT validation & accuracy
Six independent research groups have put Metric up against lab hardware — 3D motion capture systems and linear-position transducers costing thousands of dollars. In the most recent head-to-head, Metric was the most accurate VBT app tested, beating both a phone-mounted and a watch-mounted inertial sensor against a GymAware tether. We had no input into any of these studies. Here's what they found, followed by our own in-house testing.
Ranked #1 against a GymAware tether
Researchers at South China Normal University recorded 75 bench press reps with four systems running at once: a GymAware linear transducer sampling at 1000 Hz, Metric on an iPhone, Tempro on a phone accelerometer, and Spleeft on an Apple Watch. Metric tracked the tether closest — r = 0.937, ICC 0.935, with a standard error of estimate of 0.017 m/s. The two inertial-sensor apps landed at r = 0.751 and r = 0.733.
Holding up across the load spectrum
A single load only tells you so much. At the University of Zagreb, Antonio Šagovac ran 150 bench press reps at 45, 60 and 75% of 1RM against a Vitruve linear transducer: r = 0.93 mean velocity, r = 0.91 peak velocity, with intraclass correlations of 0.89 to 0.97 and coefficients of variation under 10% on every measure.
Every peer-reviewed study, in one place
Six papers from six separate research groups, tested against lab-grade hardware — 3D motion capture systems and linear-position transducers. We offered recording guidance and data export support but had no input in the published results or analysis.
- Li, Tang, Feng — South China Normal University, 2024
Metric vs Tempro vs Spleeft: bench press against GymAware
Metric most accurate of three apps · r = 0.937, ICC = 0.935 · 75 reps · bench press at 50% 1-RM vs GymAware
Read summary ↗ - Grossi et al. — University of Urbino, 2026
Metric vs Vitruve LPT: Smith machine bench press at 1-RM
Peak velocity r = 0.91, ICC = 0.888 · 18 trained men · Smith machine bench press at 1-RM load
Read summary ↗ - Šagovac, A. — University of Zagreb, 2024
Validity & reliability of Metric in the bench press
Metric v4.5 vs Vitruve linear transducer · 150 reps · r = 0.93 mean velocity
Read summary ↗ - Renner, Mitter, Baca — PLOS ONE, 2024
Concurrent validity of three smartphone VBT apps
Metric v2.3.1, Qwik VBT, MyLift vs Vicon mocap + RepOne LPT · 589 reps · mixed result for Metric — strong in the bench press, weaker in squat and deadlift, with missed and ghost reps at v2.3.1
Read summary ↗ - Taber et al. — Int. J. of Strength & Conditioning, 2023
Validity and reliability of a computer-vision system
Metric v0.5.4 vs 3D motion capture · 800+ reps · ICC mean velocity 0.79–0.98
Read summary ↗ - Trowell et al. — 2024
Validation of a commercially available mobile application
Metric v0.6.0 vs 3D motion capture
Read summary ↗
And here's what we measure in-house
The published papers above are the evidence that matters most, because we had nothing to do with them. We also run our own testing on every release. In June 2026 we compared Metric (iOS v6.9, Android v1.2) against a GymAware tether across 95 repetitions in 15 sets — bench press, deadlift, squat and cleans — deliberately including bad conditions: 45° angles, floor-level phones, 30fps footage, 720p and 4K, dim light and glare.
Rep-by-rep mean velocity agreement
Lin's concordance correlation coefficient (CCC) measures not just linear association but also systematic bias and scale differences — it's more rigorous than Pearson r for method-comparison studies. 0.9+ is considered excellent agreement; values above 0.98 approach the limit of what two instruments measuring the same physical quantity can achieve.
Rep-by-rep peak velocity agreement
Peak velocity is harder to measure accurately than mean velocity, susceptible to frame-rate and smoothing artefacts. The CCC of 0.975 across 95 reps (bench, deadlift, squat and cleans, including explosive clean peaks above 2.5 m/s) includes videos recorded at both 60fps and 30fps across Android and iOS devices.
Mean absolute error vs GymAware
MAE is the average magnitude of the per-rep difference. How far off each device is for the same repetition in metres per second. Compared to GymAware, Metric reads slightly faster on mean velocity and slightly lower on peak velocity.
More in-house validation
Additional in-house and user-contributed studies covering different reference devices, lift types, and training contexts — alongside the GymAware comparison above.
- User-contributed — June 2026
Metric vs RepOne: six real training sessions
Peak velocity CCC 0.960 · 69 reps · squat, bench, deadlift · 95–430 lb · real training data, not a lab protocol
Read summary ↗ - Metric in-house — June 2026
Metric vs GymAware: internal validation across conditions
iOS v6.9 / Android v1.2 · 17 sets · bench, deadlift, squat, cleans, jumps · per-condition breakdown across angle, frame rate, lighting and distance
Read summary ↗ - Metric in-house — 2022
Reliability & validity of the Metric VBT beta
Internal validation of an early beta. Baseline before the published peer-reviewed studies.
Read summary ↗
Use Metric in your research
Running a study and want to use Metric as a measurement tool? Get in touch. Our engineering team is available for technical questions and can help with data export.
