Force-Plate CMJ Phase Identification Errors: How Unweighting, Braking and Propulsion Get Mislabeled

Your dashboard says braking rate of force development jumped 41% between Tuesday's session and Thursday's - same athlete, same warm-up jump, same 20 kg vest. Nothing in the gym explains that. What actually changed is where the software drew the line between the braking phase and the propulsion phase on the force-time curve, and it moved because a phase-boundary rule caught the wrong slice of the trace, not because anyone got more explosive in 48 hours.

A countermovement jump has six named phases on a force plate - weighing, unweighting, braking, propulsion, flight, landing (McMahon, Suchomel, Lake & Comfort, 2018, Strength & Conditioning Journal) - but only three transition points ever need to be found automatically, and those three do not share one rule. One is a force-domain crossing. Another is a velocity-domain crossing computed by integrating the force trace over time. Mix up which rule belongs to which boundary, or let a few noisy milliseconds fool the wrong one, and every metric downstream of that boundary now describes the wrong window of time. The plate did not fail. The labeling did.

The weighing phase is roughly one second of quiet standing before the jump begins, used only to establish two numbers: the athlete's system weight, and how much natural sway sits in that weight while standing still. Everything that follows gets measured against those two numbers, which is exactly why a sloppy weighing phase causes trouble two boundaries downstream rather than announcing itself immediately.

From there, McMahon et al. (2018) define three transitions that a detection algorithm actually has to find:

• Onset (weighing to unweighting). Vertical force drops below system weight as the athlete begins flexing hips, knees and ankles. A force-domain event.
• Braking start (unweighting to braking). Force rises back up through system weight as the body starts decelerating its downward travel. Also a force-domain event.
• Propulsion start (braking to propulsion). The center of mass reaches the bottom of the dip and begins moving upward. This is a velocity-domain event - the instant integrated velocity crosses zero, not a force reading at all.

A fourth transition, takeoff, closes out propulsion when force falls back toward zero as the athlete leaves the plate. Table 1 lays out what each boundary is actually measuring and what breaks when the wrong rule gets applied to it.

The onset boundary is usually the first place things go wrong, and it goes wrong for a reason that has nothing to do with the athlete. Owen, Watkins, Kilduff, Bevan and Bennett (2014, Journal of Strength and Conditioning Research) established the threshold most force plate software still uses today: movement onset is the instant force drops by five standard deviations of the force recorded during the weighing phase, with impulse integration itself starting roughly 30 ms before that crossing to capture the very start of the deceleration. Five SD was not chosen as a round number - looser thresholds catch normal postural sway before the jump has actually started.

Barefoot, Lamont and Smith (2022, Sports) tested that assumption directly, comparing 5 SD against 4 SD and against fixed percentage-of-system-weight thresholds (2.5% and 10%) across 21 college athletes performing countermovement and squat jumps. Most derived measures held up well across methods, with ICCs between 0.79 and 0.99, but net impulse calculated off the 2.5% system-weight threshold dropped to an ICC of 0.479 - a genuinely poor reliability score sitting right next to measures scored good-to-excellent from the same trials. The threshold choice itself was the variable, not the athlete or the plate.

At scale, this shows up as a specific and common error shape. A 2026 analysis of 32,952 CMJ trials from 579 NCAA Division I athletes across 15 teams found that 45.89% of trials triggered a false-early onset on at least one of five common threshold methods, relative to a stricter rate-of-change reference, and on trials with visible pre-jump weight rocking, a common bidirectional-band method overestimated jump height by 0.943 cm. Less than a centimeter sounds trivial until it happens to be the exact size of the week-over-week change you were trying to detect.

The onset error is at least intuitive - a loose threshold catches a wobble. The braking-to-propulsion error is subtler, and more damaging, because it usually comes from applying the wrong kind of rule altogether rather than a badly tuned version of the right one.

Because the unweighting-to-braking boundary is a force-domain crossing (force rising back through system weight), it is tempting to look for the braking-to-propulsion boundary the same way - hunting for a second force-equals-system-weight crossing near the bottom of the dip. Some homemade spreadsheet analyses built along the lines of Chavda et al.'s (2018, Strength & Conditioning Journal) Excel-based approach make exactly this substitution when the true zero-velocity method feels like more work to implement. But McMahon et al. (2018) are specific on this point: braking ends and propulsion begins when center-of-mass velocity crosses zero, and velocity is not a force-plate reading at all - it is the numerical integration of net force over mass across the entire unweighting-plus-braking window.

That integration inherits every upstream error. A system weight that drifted slightly before the weighing phase was captured does not create a one-time offset - it becomes a small constant error added into every sample of the integration, and that error compounds across the roughly 400-600 ms window instead of cancelling out. A crossing near system weight found one sample late because the plate is sampling at 400 Hz instead of 1,000 Hz shifts where the integration effectively starts. None of this shows up as an obviously broken number. It shows up as a braking phase that runs 20-40% shorter or longer than it should, next to a braking RFD that looks dramatically higher or lower than the session actually produced.

Most of this is checkable without buying new hardware. Run these four checks against whatever software or spreadsheet is producing your phase durations.

1. Confirm the onset rule scales to the athlete, not a fixed number. A threshold like 10 N behaves very differently for a 55 kg gymnast than a 110 kg lineman. Five standard deviations of that athlete's own quiet-stance force, per Owen et al. (2014), scales correctly by design.
2. Require crossings to hold, not just touch. A single sample dipping below system weight is frequently noise. Reliable implementations require a crossing to persist for a minimum window - commonly in the 10-30 ms range - before accepting it as a real phase change.
3. Verify the braking-to-propulsion split actually comes from velocity. If your software cannot show you the integrated velocity curve, or the split point does not move when you perturb the velocity calculation, ask what it is actually keying off - there is a real chance it is a second force crossing wearing a velocity label.
4. Sanity-check durations against normative ranges. In trained team-sport athletes performing an unloaded arm-swing CMJ, unweighting typically runs roughly 300-500 ms, braking roughly 100-200 ms, and propulsion roughly 150-300 ms (McMahon et al., 2018). A trial reporting a 40 ms braking phase or an 800 ms unweighting phase is not evidence of an unusual athlete - it is very likely a boundary sitting in the wrong place.

A 78 kg team-sport athlete (system weight 765 N) performs an unloaded arm-swing CMJ on a 1,000 Hz dual force plate. Table 2 shows the same raw trial processed two ways: once with a robust 5 SD onset threshold and a true velocity-based braking-to-propulsion split, and once with a looser fixed threshold and a force-crossing shortcut for that same boundary - the two error types covered above, stacked on one trial.

The onset error alone stretches the reported unweighting phase from 350 ms to 430 ms and adds 80 ms to total movement time before takeoff - time the athlete never actually spent unweighting, borrowed from ordinary pre-jump sway. The boundary error is the more damaging one: shifting the braking-to-propulsion split 60 ms earlier compresses the reported braking phase from 170 ms to 110 ms, which alone inflates calculated braking RFD from roughly 6,400 N/s to roughly 9,900 N/s - a 55% jump in a metric coaches often use to flag neuromuscular readiness. Jump height itself, usually computed from flight time independent of these internal boundaries, barely moves. But reactive strength index modified (jump height divided by time to takeoff) drops from about 0.56 to about 0.50 purely because the denominator grew - an 11% fall that would read as a real decline in explosiveness on a monitoring dashboard, when nothing about the athlete's actual output changed at all.