the point is that it shouldn’t need to be compressed and packed into the level lighting data, I don’t see why it can’t be a part of a texture array that simply lives on memory.
also reflection capture processing already generates the mipmaps. yes it’s someone expensive but that’s part of the hiccup anyone would expect. and IIRC doing something like TobiGroth does already makes the proper calls and generates mipmaps, since it’s all part of the same chain.
also TobiGroth’s experiment is updating all captures at the same time. I’ve always argued this is something that could be done “in turns” capturing one at a time (based on different criteria like how close it is to the camera, and how long since it hasn’t been recaptured), and subdividing the entire capture process into smaller process chunks that could be deferred into multiple frames.
I don’t think anyone expects to update all captures every frame (unless you require a hard lighting transition, but you can hide that under a loading screen). I recon most games would need this for slow day cycles, so if we could dedicate a small portion of the frame budget (say 1-2ms) to slowly but constantly update captures over time, it could be a feature that would be functional and kept within budget.