Onset placement is the single highest-leverage decision in the signal pipeline. Every time-referenced metric on the session (both RTD windows, plateau timing, contraction duration) is computed from the onset index. An onset that is 25 ms off doesn't shift the early-window RTD by a small amount; it can invalidate the early-window slope entirely, because the 75 ms window is short enough that a quarter-window error fundamentally changes which samples you're computing across (Tillin et al. 2013).
The platform's auto-detection is built to do better than the threshold methods that dominate the older literature. When it gets a trial right (which is most of the time on a clean recording), no action is required. When it gets a trial wrong, the click-to-override workflow corrects it in one motion, and the original detection is preserved on the session record for audit.
How auto-detection works
The detector uses an adaptive threshold on the second derivative of force, refined against the local baseline noise of each individual trial. This is the operationalization of the manual-onset principles Tillin and colleagues argued for (Tillin et al. 2013): identify the last trough in the baseline-noise envelope before the deflection, ignore any drift greater than 0.5 N in the 100 ms before contraction, and use the first derivative of force as the objective fallback when the signal has been filtered.
The detector is conservative on noisy signals and aggressive on clean ones. It will refuse to declare an onset on a trial where the baseline never settled or where the rise is too gradual to be a real contraction. When it succeeds, it places the orange onset marker on the chart and downstream metrics flow from that index.
When to override
Three patterns are worth correcting by hand:
The onset marker sits too late. Usually because the rise was preceded by a small countermovement dip or a slow pre-contraction. The auto-detector waited for the unambiguous deflection above the noise envelope and placed the onset on the steep part of the rise, missing the real start. The early RTD window now opens on already-loaded muscle, and the slope is understated.
The onset marker sits too early. Less common. Usually a trial with a small false start (the patient bumped the strap, then settled, then performed the real contraction). The detector picked the false start and the real contraction is treated as a continuation. Plateau and peak still read correctly because they are post-onset window operations that don't care exactly where t = 0 is, but RTD is junk because the 0 to 75 ms window now lives inside the patient's reset, not their actual rise.
The onset marker is on a different contraction. On multi-contraction files (multiple reps in one recording), the detector returns the longest valid contraction. If that wasn't the one you wanted to look at, you'll see the marker on a different curve. Override moves it.
How the override works
Enable onset-select mode from the chart toolbar (the crosshair icon). The cursor turns into a crosshair. Click on the chart at the location where the true onset belongs. Every downstream metric recomputes from the new index immediately, the RTD bands and plateau region redraw, and the chart exits select mode automatically. The platform stores the override and the original auto-detected index on the session, so:
- The current "active" onset is your override. All metrics, all reports, all dashboard rollups use it.
- The original auto-detected index is preserved alongside it. Audit trail intact.
- A future override (yours or another clinician's) replaces the current override but does not replace the auto-detected index. The original detection is always recoverable.
When not to override
If the trial has a real problem at the onset (countermovement dip, hesitant start, false start), the right action is usually to exclude the trial and re-run, not to override the onset. Overriding sets the onset somewhere later than the auto-detection, but the early-window RTD on that trial is still going to reflect compromised tissue activation timing. See when to exclude a trial for the criteria.
The override is for cases where the contraction itself was fine but the auto-detector placed the marker in the wrong spot on the curve. It is not a way to rescue trials that should not propagate.
What to do next
- Look at the chart before you read the metrics, not after. If the orange onset marker is in a clearly different place than where you'd point with a finger, override.
- When in doubt, exclude the trial and re-run. Onset overrides are cheap but they don't fix a bad contraction.
- The override is per-trial, not per-patient. A patient who consistently shows late-detected onsets is usually a cueing issue ("push fast" not landing) or a strap-pretension issue, not a detector issue.
References
- Tillin NA, Pain MTG, Folland JP. Identification of contraction onset during explosive contractions. Response to Thompson et al. J Electromyogr Kinesiol. 2013;23(4):991-994. doi:10.1016/j.jelekin.2013.04.015