Our convolution reverb culls anything too quiet to make an audible difference during the convolution operation, which is what the dark grey section of the Impulse Response Graphic represents.
That is a pretty targeted optimization for impulse responses, and I’m not sure we could apply it more generally to other DataDrop scenarios.
Perhaps if we were to identify the dropped data as an audio file we could encode it with the rest of the audio assets instead of passing back raw bytes?