Fifteen squats, all counted "good." On rep 9 the knee was caving inward. A free rep-counter never saw it. That gap is the whole problem. Every physical-therapy app can already run free pose estimation and count reps. Counting a rep is not the same as catching the one where a recovering knee starts to buckle. We built Kinetiq to sit on the layer above the pose model. Run a post-ACL patient's home session through it and the pose stream still counts all fifteen reps, but the engine grades rep 9 differently: knee-valgus index 0.054, past the 0.03 line set for that patient's recovery profile, with tempo slowing to 1.4 as the form breaks down. Instead of one more "good" rep, it flags a re-injury risk and routes the session to a clinician before anything gets billed. (That patient is a synthetic test case we built to prove the mechanism, not a real person.) None of that judgment lives in the pose model, and it doesn't need to. Pose estimation is basically free now and only getting better. The part that decides whether a rep is safe, and whether a session is worth billing, sits above the sensor. If you already run pose estimation, what happens on the rep where the knee caves? Does anything above the rep counter catch it, or does it go in the log as one more good squat? #PhysicalTherapy #DigitalMSK #RemoteTherapeuticMonitoring #RehabTech