What gets measured gets improved. Here is exactly what MMAReader measures — and what it doesn't.
The skeleton is cosmetic; the maths is not. The silver overlay can trail fast movement by a fraction of a second — that's drawing lag, not measurement lag. Every metric is computed on each frame's own data, so the read stays frame-accurate even when the drawing plays catch-up.
Guard discipline — the share of tracked frames where your hands are home at chin height. The hand that's mid-punch is excused; the other must stay glued. Chin-height guards (shell and low-lead styles) pass; hands at the waist don't. Peekaboo targets face a stricter bar, because for that style the guard is the style.
Elbow flare — how often the elbows drift wide of the ribs. Only judged on frames where you're facing the camera enough for the read to mean something; when the angle hides it, the card says so instead of guessing.
Head movement — how much your head actually moves relative to your frame. Movement-dependent styles (Peekaboo, counter-punching) are held to a higher bar. Statues get told they're statues.
Stance width — ankle spread relative to shoulder width, measured in 3D so your camera angle can't distort it, and read only when your feet are reliably in shot.
Output — fast outward hand and foot movements per minute. Snap-backs don't count, but sharp feints and resets can, so treat it as a work-rate gauge rather than an exact strike tally. Fast footwork is not mistaken for kicks.
Film at 45°. Straight-on hides your depth; pure side-on hides your elbows. One fighter in frame, full body visible, decent light.
Known limits, honestly. Low kicks below knee height don't register yet. Two fighters in frame confuses the tracker. Elbow flare is temporarily silent while we recalibrate it for an upgraded tracking engine — a wrong number is worse than no number. Your video never leaves your device — which also means nobody here can watch it to double-check the read.