Every jump, sprint and change of direction is the same event underneath: an athlete pushing against the ground. How hard they can push, how fast, and how well they reuse the energy from landing decide a large part of how high they jump, how quickly they accelerate and how well they hold up across a season. That is what force testing measures, and for most of the history of sports science it needed a laboratory.
It no longer does. A phone filming at 240 frames a second, a tape measure and a set of published equations now recover several of the key numbers a force plate gives a coach about jumping, and sprint split times give the sprint force-velocity profile, all on the training ground. The harder question is what happens after the numbers arrive: deciding which changes are real, what they mean for this athlete in this sport, and what to do on Monday. That is where artificial intelligence earns its place, provided it is built to explain the results rather than to invent them.
What is force testing in sport?
Force testing measures how much force an athlete produces against the ground, how quickly they produce it and what that force achieves. The headline outputs are familiar to most coaches: jump height, peak and mean force relative to body mass, power, the reactive strength index (how high an athlete rebounds for the time spent on the ground), and the force-velocity profile, which shows whether an athlete is better at producing force against heavy resistance or at moving quickly against light resistance.
The tests themselves are simple and well known:
- Countermovement jump. A dip and jump with hands on hips. The standard measure of lower-body power and the most used check of neuromuscular readiness.
- Squat jump. The same jump from a held squat, with no dip. Compared with the countermovement jump it shows how well an athlete uses the stretch-shortening cycle.
- Drop jump and the 10-5 repeated jump test. Measures of reactive strength: how quickly and stiffly an athlete rebounds from the ground.
- Loaded jumps. Jumps at four to six loads, which give the jump force-velocity profile.
- Sprint splits. Times at several distances over 30 metres or more, which give horizontal force, maximal velocity and power in sprinting.
- Single-leg jumps. Left and right compared, to look at limb differences.
Why does force matter for athletes and coaches?
Force is the common currency of speed and power. In a study of 100 metre sprinting, Morin and colleagues concluded from correlations in 13 runners that the main mechanical determinants of performance were a velocity-oriented force-velocity profile, applying force in a forward direction during acceleration, and a higher step frequency from shorter ground contacts (Morin et al., 2012). The sample was small and mostly non-specialists, so treat it as a strong pointer rather than proof: how an athlete applies force may matter as much as how much force they have.
Force testing also tells a coach how an athlete is coping. A meta-analysis of 151 articles found the countermovement jump to be a sensitive marker of neuromuscular fatigue and supercompensation, and that the average of several jumps was more sensitive than the best single jump (Claudino et al., 2017). That is a useful, cheap signal in a busy training week.
Finally, the profile is individual. Across 666 elite athletes from 23 sports, sprint force-velocity profiles varied more between individuals than between sports, with large overlap and large differences among athletes who performed equally well (Haugen et al., 2019). Two sprinters with the same 30 metre time can get there in very different ways, and they may need different training.
Can phone video replace a force plate?
For jump height and sprint mechanics, phone video comes close enough to be useful, and the same timing principles apply to reactive strength. Two conditions apply: the frame rate must be high, and results should only ever be compared with results measured the same way.
The flight-time method calculates jump height from how long the athlete is in the air, and filming lets a coach mark the take-off and landing frames exactly. A meta-analysis of 21 studies with 839 participants found no significant difference between the My Jump app, which uses frame marking, and criterion systems, with a pooled ranking correlation of 0.989 and near-perfect reliability, although results varied between studies (Gençoğlu et al., 2023). Frame rate matters: timing error in flight time fell from 3.4 ms at 120 frames a second to 1.8 ms at 240 (Pueo et al., 2023), and in 24 active men filmed on an iPhone at 240 frames a second, flight-time heights did not differ significantly from the force plate when they landed on straight legs (Nishioka et al., 2025). By contrast, an AI app estimating height automatically from 30 frame-a-second video underestimated jump height by about 3 cm, with wider error (Ríos-Gallardo et al., 2026).
There is a known bias to respect. Flight time reads slightly higher than the force plate’s impulse-momentum method: about 2.5 cm on average in one recent study, even though both methods were highly reliable (Hobein et al., 2026), and by 2 to 6% for unloaded jumps across a systematic review of 16 papers (Eythorsdottir et al., 2024). The practical rule is simple: flight-time heights are excellent for tracking an athlete over time, but they should never sit in the same table as force-plate heights.
For sprinting, split times are enough. The simple method of Samozino and colleagues estimates horizontal force, velocity and power from a sprint’s split times and agreed with force plates to within 5% (Samozino et al., 2016). A replication on a 50 metre force-plate track found an average bias of about 5%, with repeatability similar to the force plates themselves (Morin et al., 2019), and an iPhone app that times splits from slow-motion video agreed closely with photocells and radar in 12 trained sprinters (Romero-Franco et al., 2017). Setup still matters: timing that ends at 20 metres or earlier overestimates maximal force, so the whole acceleration should be covered, and for team sport athletes gates at 10, 20 and 30 metres are enough (Haugen et al., 2020).
What makes video force testing reliable?
Reliability comes from a few unglamorous details, and each one can be built into the tool rather than left to memory:
- Frame rate from the file, not the label. A video’s own timestamps say exactly when each frame was captured. Reading them directly avoids the small but real errors that come from assuming a steady frame rate.
- 240 frames a second for timing. iPhones and many Android phones record this in their slow-motion mode; some record only 120. Film with the phone’s camera app in slow motion and upload the clip, because recording inside a browser is often limited to 60 frames a second. From 120 to 239 frames a second results carry wider error bounds, and below 120 they should be treated as indicative only.
- The same protocol every time. Hands on hips, the same dip depth, landing on straight legs with toes pointed, and the same warm-up. Bent-knee landings lengthen flight time and inflate height.
- Measured inputs. Force and power from jumps need body mass and push-off distance (the leg’s extension range). Changing push-off distance changes the force-velocity profile (Pommerell et al., 2025), and so does the way it is measured (Paulsen et al., 2025), so it has to be measured, and measured the same way.
- The same timing setup for sprints. Never mix timing systems in one athlete’s history.
How do you know whether a change is real?
Every test has day-to-day noise, and the most common mistake in athlete monitoring is reacting to it. The standard approach is to compare any change with the test’s typical error, the normal variation in an athlete’s repeated results (Hopkins, 2000), and with the smallest worthwhile change, the smallest difference that matters in practice. A common working convention for the smallest worthwhile change is 0.2 of the spread between athletes in the squad.
In practice this means a coach should know two numbers before acting on a result: how much this athlete’s score normally moves from session to session, and how big a change would actually matter. A 1.5 cm drop in jump height is meaningful for one athlete and noise for another.
Where does AI help, and where must it stay out?
Collecting force data is now the easy part. The bottleneck is interpretation: turning a set of numbers into a clear, athlete-centric message a coach can act on, in the language of their sport. That is a job well suited to AI, as long as the system is designed around three rules.
First, the calculations are not the AI’s job. Jump heights, forces, reactive strength and profiles should come from published, versioned equations that give the same answer every time. A language model should never compute a number a coach will rely on.
Second, the AI explains findings that rules have already made. A rules engine decides what the data says: this change is bigger than the athlete’s typical error; this profile is unclear and needs a retest; this asymmetry is within the test’s noise. The AI’s role is to phrase those findings for the athlete in front of the coach, connect them to the athlete’s sport and recent results, and suggest options, with every number in its wording checked against the calculations before the coach sees it.
Third, the guidance is personal, not generic. Athlete-centric guidance means judging each result against the athlete’s own baseline rather than a population norm, framing it for their sport, and saying honestly how sure we can be. A 400 metre runner and a padel player with the same countermovement jump result need different conversations.
The research is clear about where AI, and coaches, should hold back:
- Force-velocity profiles are not prescriptions. A meta-analysis found the evidence that training based on the jump profile beats non-optimised training to be weak (Solberg et al., 2025). The profile’s slope varied 14 to 30% between days even with near-perfect fits (Lindberg et al., 2021), and over 80% of 16 female handball players changed category depending on the calculation method used (Paulsen et al., 2025). A good system treats the profile as one input to a conversation and says “unclear, retest” when the noise band straddles the line.
- Asymmetry has no universal threshold. A narrative review drawing on ten years of data collection found no reasonable scientific basis for a blanket 10% threshold and advised judging asymmetry against the test’s own variability (Bishop, Afonso and Turner, 2026).
- Jump tests do not predict an individual’s injury. Week-to-week drops in countermovement jump performance have been associated with more lower-limb injuries across groups of young players (Sastre-Munar et al., 2026), but that is a reason to check in with an athlete and review training load, not a risk score.
What does athlete-centric guidance look like in practice?
Take an illustrative example. A sprinter’s mean countermovement jump has fallen by more than her typical error for two sessions in a row, while her drop jump reactive strength has held. A generic report would say “jump height down 6%”. An athlete-centric one says that the drop is bigger than her normal variation, that it has lasted two sessions, that her reactive qualities are unchanged, and that the sensible next step is to review the last fortnight’s load and retest after an easier day, rather than to change her programme. It cites the evidence behind each statement and does not claim more certainty than the data allows.
The same principle applies to good news. A rise in reactive strength in a youth athlete might be training, or it might be growth. Saying so is more useful to a coach than a confident headline.
How does Athleet.Ai’s Coach Force Lab put this together?
Coach Force Lab is the force testing tool in the coach editions of Athleet.Ai. Coaches film on a phone or upload a clip, mark the take-off and landing frames on the exact frame shown, and get jump, reactive strength, force-velocity, sprint and asymmetry results calculated with published methods. Each result carries an evidence grade, and each timed result shows how far it could move from frame timing alone. Video stays on the coach’s device.
The written coaching report follows the rules above: a rules engine decides the findings, AI phrases them for the athlete and the sport, and every number in the report is checked against the calculations before the coach sees it, with the rules’ own wording used if any check fails. Changes are judged against the athlete’s own typical error once enough sessions exist. Inside the tool, a science page sets out the sources behind every calculation, what each means for coaching, where the tool makes its own design choices, and a daily feed of new research.
Where the evidence stops
Video-based force testing is well supported for jump height and sprint mechanics in controlled conditions, and promising for reactive strength. It is less settled for using force-velocity profiles to prescribe training, for limb asymmetry and for any link to injury. Most validation studies are small and many involve trained adult men, so results in youth, female and para athletes deserve extra care. The honest position is to use these tests to track the qualities you are training, judge each athlete against themselves, and keep the coach in charge of the decision.
Frequently asked questions
What is force testing in sport?
Force testing measures how much force an athlete applies to the ground, how quickly, and what it achieves. Common tests are the countermovement jump, squat jump, drop jump, 10-5 repeated jump test, loaded jumps for a force-velocity profile, sprint splits and single-leg jumps.
Can a phone replace a force plate?
For jump height and sprint split times, phone video at 240 frames a second agrees closely with laboratory systems. Flight-time heights read slightly higher than force-plate impulse heights, so compare results measured the same way and keep the methods separate.
What frame rate do I need for jump testing?
240 frames a second. Flight-time error roughly halves from 120 to 240 frames a second. Many phones, including iPhones, record 240 in slow-motion mode. From 120 to 239 frames a second results carry wider error bounds; below 120 treat them as indicative only.
How often should athletes do force testing?
A countermovement jump once a week, on a consistent day relative to competition, is a common monitoring routine. Fuller profiles such as loaded jumps or sprint force-velocity profiles are typically repeated each training block, often every six to eight weeks.
Can AI predict injuries from jump tests?
No. Drops in jump performance are associated with more injuries across groups of athletes, but no jump test has been shown to predict an individual’s injury. A drop beyond an athlete’s normal variation is a reason to review training load and talk to the athlete.
How should AI be used with force testing results?
AI should explain results, not calculate them. The numbers should come from published equations and the findings from transparent rules; AI can then turn those findings into clear, athlete-specific guidance in the language of the sport, with every number checked against the calculations.
Coach Force Lab is available in the coach editions of Athleet.Ai. See how it works or explore the editions.