After effects generative fill analyse and render times

You misunderstand, I think. You don't track the mask. A conventional old style corner pin should work just fine. The white badge has four clearly visible edges and it should be easy enough to just track a white solid and presumably even faking the shading with a simple gradient would work. Likewise, since all the text on the badge appears to be black, simply extracting it with Shift Channels or Channel Combiner to generate a matte/ inverted white fill should work. This could really be done in multiple ways and they're all more user friendly than relying on a complex fill algorithm. The only sticking point is that you still have to tackle this in manageable segments of around three to five minutes at most and make sure they then blend smoothly, but otherwise this is probably two or three hours of work and then it's off to render. No need to waste another week.

Mylenium 

Long and short: You probably need to use a conventional masking workflow and patch things up with tracking. Provide a reference image and people can tell you what to do. Your whole workflow simply sounds wrong.

Mylenium