I'd be interested to know whether way and useful is the main bottleneck here. A quick controlled test could answer that pretty fast. The part about "A useful way to understand AI Motion Capture is to separate the goal, the input, the model or tool, and the way the out…" is the bit I would test...