Some exercise apps show a score after each repetition: 82 out of 100, say. It looks precise, and the decimal-free number invites you to treat it like a grade. It is worth knowing what sits behind it, because a score like this is a comparison, not a diagnosis, and it is only as good as the thing it compares you to.
Three ways software judges a movement
A 2020 review in Computers in Biology and Medicine grouped the computational approaches into three families. Discrete scoring systems learn from examples that clinicians have already rated and try to reproduce those ratings. Rule-based systems check specific conditions, such as whether the knee reached a set angle or the trunk stayed within a tolerance. Template-based systems compare your whole movement against a reference recording and measure how far apart the two are. Most real products blend these: a few hard rules for safety, a template for the overall shape, and a learned model to turn the differences into a single number.
What the number is compared to
Whichever method is used, the score is relative to something: a clinician's ratings, rules the therapist chose, or a reference performance. That reference is usually built from healthy people doing the exercise well. For someone six weeks after knee surgery, the reference may be unreachable by design, and a 60 can represent excellent work for where they are. This is why researchers in the field argue that the comparison should adapt to the individual and move as they recover, rather than holding everyone to the same template.
What drags a score down that isn't your fault
- Camera angle. Joints on the far side of the body are estimated less accurately than those facing the camera, and partly hidden limbs introduce error that looks like poor form.
- Lighting and clothing. Shadows, backlighting, and loose or dark clothing shift where the software places your joints.
- Drifting out of frame. If a foot or hand leaves the picture, the model guesses, and guesses score badly.
- Speed. Very fast repetitions can outrun the frame rate, so the software never sees the top or bottom of the movement.
- Plain model error. Single-camera joint angles can be off by several degrees even under good conditions, which is enough to move a score by a few points from one rep to the next.
How to read your own scores
- Watch the trend across sessions, not the number on one rep.
- A sudden drop after you changed rooms, lighting, or clothing is a setup problem first. Check the camera before you doubt yourself.
- A steady decline over several days, especially alongside rising pain, is worth a message to your therapist.
- A score is not pass or fail. It cannot detect injury, and it cannot tell whether an exercise is appropriate for you today.
How KineTrue fits
KineTrue records a 0 to 100 quality score per exercise within a session, next to reps or hold time, range of motion, and your 0 to 10 pain rating. Your therapist sees these together. A quality score falling while pain climbs tells a different story from one falling while pain stays flat, and that pairing is what the therapist reads, not the score alone.
A high score means your movement matched the reference closely. It does not mean the exercise is right for you today. If something hurts in a new way, stop and ask, whatever the number says.