Metric vs RepOne: a real-world accuracy comparison across six training sessions

Most VBT validation studies happen in a lab: one session, controlled protocol, a researcher managing the setup. This one is different.

Joe ran Metric and a RepOne linear position transducer simultaneously across six of his actual training sessions over two weeks in May and June 2026. Squat, bench press, deadlift. Loads from 95 to 430 lb. No special setup — he just had both devices running on the same sets. He shared the data with us and gave permission to publish the analysis.

69 matched reps made it into the study after six reps were excluded for confirmed data quality reasons (more on that below). This is what the numbers show.

What is RepOne?

RepOne is a tether-based linear position transducer — a cable attached to the barbell that physically extends and retracts to measure vertical displacement, from which velocity and ROM are derived. It samples at high frequency and is widely used by strength coaches and research labs as a trusted VBT reference device.

Headline results

Agreement is measured using Lin’s Concordance Correlation Coefficient (CCC), which combines both precision (how tightly the numbers track together) and accuracy (how close they are to the 45° agreement line). Scale: 0.99+ near-perfect · 0.95–0.99 substantial · 0.90–0.95 moderate · below 0.90 poor.

Metric	n	Pearson r	CCC	Bias (Metric − RepOne)	MAE
Mean velocity (m/s)	69	0.926	0.866	+0.076	0.091 m/s
Peak velocity (m/s)	69	0.968	0.960	+0.044	0.076 m/s
Range of motion (cm)	69	0.939	0.862	+1.9 cm	3.2 cm / 6.7%

Peak velocity is strong across all three lifts — r = 0.968, CCC 0.960 — from 95 lb warm-ups to 430 lb singles. Mean velocity on squat and bench is equally solid. On deadlift the two devices diverge, and the reason traces directly to where each one measures the bar, which turns out to be the most interesting thing in the dataset.

Bench press — near-perfect agreement

Metric	n	Pearson r	CCC	Bias
Mean velocity	6	0.995	0.980	−0.022 m/s
Peak velocity	6	0.999	0.959	−0.054 m/s
ROM	6	—	—	−2.2 cm / 7.0%

Bench press is about as close to perfect as real-world data gets: r = 0.995 mean velocity, r = 0.999 peak velocity. Metric reads fractionally lower than RepOne on bench (−0.022 m/s mean, −0.054 m/s peak) — consistent and small. ROM within 7%.

Squat — very strong

Metric	n	Pearson r	CCC	Bias
Mean velocity	21	0.974	0.949	+0.049 m/s
Peak velocity	21	0.977	0.909	+0.099 m/s
ROM	21	—	—	4.6 cm / 8.3%

r = 0.974 mean velocity, r = 0.977 peak velocity — CCC 0.949 and 0.909 respectively. Metric reads +0.05 m/s above RepOne on squat mean velocity across all loads, a consistent offset that holds through every session. Peak velocity agreement is tight from 135 to 375 lb. ROM absolute agreement is 8%.

Deadlift — peak excellent; mean velocity tells a physics story

Metric	n	Pearson r	CCC	Bias
Mean velocity	42	0.924	0.813	+0.103 m/s
Peak velocity	42	0.984	0.978	+0.031 m/s
ROM	42	—	—	2.7 cm / 5.8%

Deadlift peak velocity: r = 0.984, CCC 0.978 — on par with controlled lab studies. ROM absolute agreement is the best of any lift at 5.8%.

Mean velocity is where the two devices diverge, and the root cause is not a sensing error. It is physics.

Why deadlift mean velocity diverges: bar bend

RepOne’s tether attaches to the centre of the bar. A barbell under load bends — the centre rises before the plates leave the floor. RepOne therefore opens its concentric timing window the moment the bar bends, which at heavy loads can be well before the plates actually move.

Metric tracks the plate/sleeve. It starts timing when the load actually breaks the ground.

The result: RepOne measures a longer concentric duration on heavy deadlifts, which produces a lower mean velocity — and Metric’s plate-based reading looks relatively high by comparison. Joe’s data shows this cleanly:

Load	n	RepOne duration	Metric duration	Window gap	Mean-vel gap
135 lb (light)	12	0.68 s	0.69 s	−0.01 s	+0.07 m/s
~225 lb	5	0.93 s	0.86 s	+0.07 s	+0.11 m/s
275–345 lb	11	1.48 s	0.96 s	+0.51 s	+0.16 m/s
405–430 lb	13	2.05 s	1.29 s	+0.76 s	+0.11 m/s

At 135 lb the concentric windows are identical — bar bend is negligible at light load. At 405–430 lb, RepOne opens its window 0.76 seconds earlier than Metric. That extra time is bar deformation, not bar travel. The correlation between load and RepOne’s excess window length is r = +0.84 — it scales precisely with weight.

Which measurement is more correct? Metric’s. Mean velocity is supposed to reflect the effort of the actual lift — bar flex time before the load breaks the ground is not part of that. The plates-moving window captures what the body did. Both devices are internally consistent and valid for tracking the same lifter over time; you just need to know which window each one is using when comparing numbers between them.

When the load-scaling effect is accounted for, deadlift mean-velocity Pearson r rises from 0.84 to 0.95.

Data quality: what was excluded and why

Six reps were excluded entirely from the analysis — not selectively by metric, but as whole reps, so per-metric sample sizes stay consistent.

1 rep: squat half-rep (5/22, 300 lb). Metric detected only the first portion of the squat: ROM 30 cm vs RepOne 57 cm, peak velocity 1.67 vs 0.54 m/s. The numbers are directionally opposite — a clear false positive detection. Excluded flag: excluded_halfrep.

5 reps: deadlift bar-path tracking artifact. On five deadlift reps, Metric’s recorded concentric duration was 2.2–3.3× longer than expected, halving mean velocity, while peak velocity and ROM remained accurate. Investigating the bar-path telemetry, the cause is a synthetic “draw-in” line — a slow, perfectly linear vertical ramp that the tracker generates before the actual pull, which then blends into the real rep start. Each row below is a sequence of bar velocities (m/s) sampled frame-by-frame. The │ marks where Metric’s rep-start threshold was crossed and the concentric window opened — everything left of it is before the window, everything right gets averaged into mean velocity. Two 315 lb deadlifts:

                           ← before window opens ──────────── │ ── concentric pull →
clean  315 lb (mv 0.52):  0.00  0.00 -0.01 -0.02 -0.00  0.03 │ 0.33  0.55  0.84
glitch 315 lb (mv 0.26):  0.09  0.11  0.13  0.15  0.18  0.19 │ 0.33  0.68  0.82

The actual pull (right of the │) is identical on both reps. The glitch rep carries a phantom pre-ramp (~0.1→0.2 m/s, dead-straight) in the frames before the window opened. That ramp crossed Metric’s rep-start velocity threshold, opening the concentric window 2–3 seconds early and diluting the mean. This is a bar-path generation bug, not the rep segmenter — and it is now fixed.

The artifact connects directly to one of Metric’s design choices: it tracks from the moment you start recording, with no manual rep-start button required. That always-on readiness is what surfaced the phantom ramp — on heavy slow singles, the tracker was already watching bar micro-motion before the pull began, and occasionally generated a draw-in line to bridge that stillness into the rep start. Finding it required exactly the kind of multi-session real training data Joe provided. With enough reps at enough loads across enough sessions, the pattern was unambiguous and the fix was clean.

These exclusions are honest and conservative. The remaining 69 reps are clean by every metric.

What this means in practice

For coaches and athletes using Metric alongside a RepOne (or similar tether LPT):

Peak velocity is directly comparable across squat, bench and deadlift. r = 0.968 across all lifts, 0.984 on deadlift alone. Use it to track explosiveness, power output, and fatigue.

Mean velocity is comparable on squat and bench. r = 0.974 squat, r = 0.995 bench, with consistent offsets (+0.05 m/s squat, near-zero bench) that are stable across loads and sessions — fully usable for load-velocity profiling and autoregulation.

Deadlift mean velocity reads higher than RepOne at heavy loads because Metric measures from the plates, not the bar centre. The gap grows predictably with load. When comparing to RepOne deadlift MV reference values at 80%+ 1RM, apply the offset — it is load-scaled and consistent, not noise.

ROM absolute agreement is 6–8% across all lifts — solid for a secondary quality check.

Thanks to Joe

This analysis exists because Joe tracked both devices through a real training block and shared the data. This kind of real-world comparison — actual programming, actual loads, multiple weeks — is harder to fake than a lab protocol and arguably more relevant to how coaches and athletes actually use the tools.

Beyond the accuracy numbers, Joe’s data led directly to a bug fix. The tracking artifact on heavy deadlift singles was not something we would have caught in a controlled session — it took repeated heavy slow singles across multiple weeks to generate enough cases to isolate and reproduce it. Joe flagged the anomalous reps, shared the telemetry, and the fix shipped. That’s the value of real training data from real athletes.

If you are running both Metric and a tether LPT in your training and want to share the data for analysis, get in touch.

The full rep-by-rep dataset and analysis methodology are available on request.