The plateau region of a maximum-effort isometric contraction is where three different control questions get asked of the same signal. The patient was instructed to hold their peak. How far below peak did they actually settle? That's Drift from peak (nRMSE). Once they settled, how much did their force vary? That's Steadiness (CV). Did that variation move smoothly, or did it jerk? That's Smoothness (Yank). Together they make up Control on the dashboard: how well force is held, alongside Strength (how much) and Speed (how fast it builds). Each tile leads with the plain name and shows the technical one beneath it.
The three are not redundant. Reading them as a set is what the recent measurement-theory literature argues for, and it's the reading the platform's metric tiles are designed for (Yacoubi & Christou 2024; Sherman et al. 2024). The remainder of this article walks each one in turn, what it means physiologically, and where the three diverge.
Drift from peak
How far force sags from its best while holding. Lower is better.
nRMSE
Normalized Root Mean Square Error from peak, expressed as a percentage.
For each sample in the plateau window, the platform takes the deviation from the peak force value (always negative or zero, since the plateau sits below peak by definition), squares it, averages, square-roots, and divides by peak. The number is a dimensionless percent.
Mechanically:
nRMSE = sqrt( mean( (force_i − peak_force)² ) ) / peak_force × 100
Conceptually: "on average, force during the plateau deviated X percent from the patient's own peak." It's the metric you read when you want to know how well the patient sustained their own maximum. A patient whose curve hits 600 N at peak but drops to 540 N average across the plateau has an nRMSE around 10%. A patient whose curve hits 600 N peak and sits at 590 N average across the plateau has an nRMSE around 1.7%.
Why normalize to peak rather than mean? Because the clinical question is "did the patient maintain their maximum," not "how variable was the signal around its own average." A patient who fell off sharply has high nRMSE because the deviation from peak is large, even if the signal during the fall-off was perfectly smooth. nRMSE captures the fall-off magnitude, which is the clinical thing of interest on a maximum-effort hold.
Healthy adults during short MVICs typically produce nRMSE between 2 and 10%. Values above 15% suggest meaningful force decline during the plateau. The metric is dimensionless and unit-invariant.
Steadiness
How much force wavers during the hold. Lower is steadier.
CV
Coefficient of variation across the plateau, expressed as a percentage.
CV = (SD / mean) × 100
Where SD is the standard deviation of plateau force and mean is the mean plateau force. The conceptual difference from nRMSE: CV asks how variable the force was around its own average, regardless of what that average was. A patient whose plateau drops 10% off peak and then sits there with very little variation has a low CV (because the variation around the 540 N mean is small) but a moderate nRMSE (because the 540 N is meaningfully below 600 N peak).
This is the classic force-steadiness measure. The published literature on force control predominantly uses CV, and the published normative ranges are in CV. Force steadiness reflects the common synaptic input to the motoneuron pool: slow common oscillations under 10 Hz in motor unit discharge times drive the force fluctuations CV captures (Enoka & Farina 2021). The same review notes CV is force-dependent: it's higher at low target forces and falls to a relatively constant value at moderate-to-high force. The platform computes CV on the plateau of a maximum-effort contraction, so the force-dependence is largely controlled by protocol.
CV is also moderately associated with functional outcomes: in studies of grooved pegboard performance, CV at 5 to 10% MVC explains 36 to 47% of variance; in 6-minute walk tests, CV in plantar- and dorsiflexors explains 36 to 54%; in postural sway, CV at low target forces explains 14 to 69% across balance conditions (Enoka & Farina 2021). The mechanisms behind variation in CV are reasonably well characterized.
Healthy adults during sustained MVIC typically produce CV between 2 and 8%. Values above 15% suggest significant force unsteadiness.
Smoothness
How jerky the small changes in force are. Lower is smoother.
Yank
Root mean square of the first derivative of force during the plateau, amplitude-normalized to peak force. Units: s⁻¹ (dimensionless).
Yank = RMS( dF/dt ) / peak_force
The conceptual difference from CV: CV captures how much the signal varies. Yank captures how abruptly it varies. A signal that drifts slowly up and down around its mean has low Yank. A signal that jerks rapidly around its mean has high Yank, even if the amplitude of the variation is the same.
This distinction matters because CV and Yank can move in opposite directions. A patient with a tremor superimposed on a steady hold has a higher Yank than the same patient without the tremor. The CV may stay roughly the same, or even decrease, because the tremor adds high-frequency oscillation without necessarily increasing the standard deviation around the mean. The dramatic example: essential tremor patients show CV roughly 37% higher than healthy controls, but Yank around 250% higher (Yacoubi & Christou 2024). CV alone would miss most of the tremor; Yank surfaces it.
The amplitude normalization is essential. The original proposal for Yank as a measure used the raw RMS(dF/dt) value in N/s, but that form scales with movement amplitude and cannot be compared across patients or target forces (Sherman et al. 2024). Dividing by peak force converts Yank into a dimensionless rate (s⁻¹) that can be compared across patients, sessions, and joints. The platform's Yank implementation uses the amplitude-normalized form for exactly this reason. The clinical and theoretical conversation continues, but the normalization is what makes the metric usable as a population-comparable measure today.
Yank is the newest of the three. The post-ACL force-control meta-analysis specifically suggests that nonlinear / regularity-style methods detect more and larger group differences than linear CV (Schwartz et al. 2025), which is the empirical case for not relying on CV alone.
Reading the three together
A useful frame: Drift from peak asks did they sustain it, Steadiness asks was it consistent, Smoothness asks was it smooth. The three patterns most worth recognizing on a session:
- High nRMSE, low CV, low Yank (drifts, but steady and smooth). The patient dropped off peak and then held steady at a lower force. The variability and smoothness are fine; the issue is the magnitude of the fall-off. Often a fatigue or motivation question.
- Low nRMSE, high CV, low Yank (near peak, but wavering). The patient stayed near peak on average but their force oscillated slowly around the mean. The smoothness is fine; the issue is consistent output. Often a motor control or attentional question.
- Low nRMSE, low CV, high Yank (near peak and steady, but jerky). The patient held a high mean force with a small amplitude of variation, but that variation moved abruptly. The magnitude and consistency are fine; the issue is smoothness. Common in patients with tremor, with neurological involvement, or with very fresh post-op irritability.
None of these patterns is exclusive. They show up as combinations, and the LSI radar is where the asymmetry shape becomes legible. A patient with high Yank on both sides but a symmetric Smoothness LSI is a different clinical picture than a patient whose asymmetric Yank drags Smoothness LSI below 90%. In the dashboard's simplified view the three are combined into a single Control reading, with the lowest of the three named beneath it; switch the patient to the expanded view to see each one on its own spoke.
What about the steadiness reliability flag?
When the platform detects that the device's quantization step or transition rate is limiting how much of the steadiness signal reflects real motor control versus hardware resolution, it raises a reliability flag on the Control tiles. The flag is device-agnostic by design (it's computed from signal properties alone), and it is the Control-side equivalent of the RTD reliability badge. See the steadiness reliability flag for the details.
Try it in the Dynamometry Lab: see Drift from peak, Steadiness and Smoothness (nRMSE, CV and Yank) on three simulated holds, from steady to shaky and fading.
What to do next
- Read the three Control tiles together, not one at a time. The diagnostic content is in the combination.
- For longitudinal tracking, the absolute values matter less than the trajectory. A patient who started at nRMSE 22% and CV 16% and is now at nRMSE 13% and CV 10% has made meaningful progress even though neither number is back inside the typical range yet.
- For cross-patient comparison, Smoothness (Yank) in particular is most credible when the protocol is consistent. Different target percentages, different muscles, different rates all change Yank's interpretation more than they change CV's.
References
- Enoka RM, Farina D. Force steadiness, from motor units to voluntary actions. Physiology (Bethesda). 2021;36(2):114-130. doi:10.1152/physiol.00027.2020
- Yacoubi B, Christou EA. Rethinking force steadiness, a new perspective. J Appl Physiol. 2024;136(5):1260-1262. doi:10/gtvkp7
- Sherman DA, Darendeli A, Soto O, et al. Beyond force steadiness, potential challenges in measuring smoothness of force through yank. J Appl Physiol. 2024;136(5):1266-1267. doi:10/gtvpr5
- Schwartz AL, Koohestani M, Sherman DA, et al. Knee extensor and flexor force control after ACL injury and reconstruction, a systematic review and meta-analysis. Med Sci Sports Exerc. 2025;57(2):238. doi:10.1249/MSS.0000000000003574