The steadiness reliability flag is the steadiness-side equivalent of the RTD reliability badge. It fires when the platform detects that the device's hardware resolution is limiting how much of the variation in the steadiness metrics reflects the patient's real motor control versus the device's own measurement floor.
Like the RTD reliability flag, it's device-agnostic by design: the platform infers the limit from properties of the signal itself, not from a device whitelist or a metadata header. If the platform can't tell apart 0.5 N of real fluctuation from 0.5 N of measurement noise, the flag goes up regardless of which device produced the signal.
What the flag is measuring
Two underlying mechanisms can drive steadiness metrics that don't reflect physiology:
Quantization step. Some devices report force values in fixed increments. A device that reports in 1-pound increments rounds the actual force to the nearest pound, which is about 4.45 N. On a plateau averaging 100 N, that quantization step is about 4.5% of the mean, which is right in the middle of the range where a healthy CV value would naturally sit. The reported CV in that scenario is dominated by the rounding step, not by the patient's motor control.
The platform measures the smallest non-zero difference between consecutive samples on the pre-filter signal and converts it to a percentage of the plateau mean. If that percentage approaches or exceeds the reported CV, the flag goes up.
Effective transition rate. Some devices report at a high nominal sample rate but only produce a new distinct force value every few samples (the signal sits flat at the same rounded value for a while, then jumps to the next). On a 100 Hz signal with only 5 distinct values per second, Yank specifically becomes meaningless because Yank is computed from the time derivative of force, and the time derivative of a step function is zero between steps and infinite at each step.
The platform counts distinct-value transitions per second on the pre-filter plateau signal. If the transition rate is too low to support the metric being computed, the flag goes up.
The three tiers
The flag uses the same three-tier vocabulary as the RTD reliability badge:
- High (no badge): the device's resolution is fine enough that the steadiness metrics are dominated by patient physiology rather than measurement floor.
- Moderate (yellow): the resolution is enough to compute a meaningful steadiness metric, but interpreting a small change between sessions is harder than usual because the measurement floor is closer to the value being measured.
- Low (red): the resolution is comparable to or larger than the value being measured. The steadiness metrics still ship (no metric is suppressed), but they should not be interpreted as reflecting motor control.
The badge appears on whichever of the steadiness tiles (nRMSE, CV, Yank) is most affected. In practice, Yank is the most sensitive to transition-rate limits and CV is the most sensitive to quantization limits. nRMSE sits between them: it's affected by both but usually less severely than either CV or Yank.
Why this isn't a sample-rate flag
The RTD reliability flag is essentially "how many samples landed in your time window." The steadiness flag is different. A device can stream at 200 Hz with low effective resolution (lots of samples, but most of them are the same rounded value), and the RTD flag would read high while the steadiness flag goes up. The two flags answer different questions about the same signal.
The opposite case is also possible: a device streaming at 40 Hz with very high resolution can produce a low RTD reliability badge (because the RTD window holds only three samples) while the steadiness flag reads high (because each of those samples is precise). Cover both flags before drawing a conclusion from any of the metric tiles.
Reading the flag in practice
The flag changes how the steadiness numbers should be read, not whether they're shown:
- At low confidence on a within-patient longitudinal track, the trend direction is still usually informative. The device's resolution floor doesn't change between visits, so if a patient went from CV 8% to CV 4%, the direction is real even when the absolute floor is in question. Don't read absolute values; do read trajectories.
- At low confidence on a between-side LSI comparison, the asymmetry is still legible. The same device measured both sides; the resolution floor cancels in the ratio. A 70% steadiness LSI is meaningful even when the absolute CV values it's computed from are quantization-limited.
- At low confidence on a cross-patient comparison or a comparison to published normative values, hold off. The literature's reference ranges assume measurement floors well below the values being reported. Comparing a low-confidence CV to a published threshold is comparing two things that aren't on the same scale.
What to do next
- Check the badge before reading the steadiness numbers, not after. Same rule as the RTD reliability badge; the tier is the context the values sit in.
- If the badge appears consistently for a specific patient, the issue is the recording method, not the patient. Switching dynamometers or reconnecting may help; sometimes the issue is a firmware mode on the same device that's reporting in coarser increments than the underlying sensor is capable of.
- Steadiness numbers from a low-confidence session are still worth keeping on the record. They support trajectory tracking and LSI comparisons even when they don't support absolute interpretation.
For the metric definitions themselves, see steadiness, reading nRMSE, CV, and Yank together.