The pose model needs landmarks
Strikt reads a skeleton from the video frame: shoulders, elbows, wrists, hips, knees, ankles. Each landmark comes with a confidence. When confidence is high the counter is certain; when it drops, the counter abstains until it recovers.
That is the whole design principle: uncertainty pauses the count, it never refuses a rep.
Things that lower confidence
Most of them are about the frame, not about you.
- Body parts out of frame: feet in a frontal push-up, hands overhead in a jumping jack.
- Strong backlight, a window behind you, or a dark room.
- Loose clothing that hides the joints.
- A phone so close that the movement leaves the picture at the bottom of the rep.
- Another person moving through the frame.
What the setup stage checks
Before a set starts the viewfinder shows four corners and one instruction: get your whole body inside. Any phone height works. When the skeleton is stable the countdown starts on its own. That is the only gate, and it is there so the set is not interrupted later.
Angles
Every movement counts from the front, the side, or anything in between. The angle written in the guide for each movement is where the camera reads the most technique detail, not a requirement for the count. If your room only allows one angle, use it.