That's a limitation. Temporal supersampling, both DLSS v2+ and others, work by sparsely sampling the scene and combining jittered offsets over multiple frames,

That's a limitation. Temporal supersampling, both DLSS v2+ and others, work by sparsely sampling the scene and combining jittered offsets over multiple frames, but in fast moving stuff with big foreground/parallax elements there are large sections of the screen that have only been disoccluded for a single frame, so you only have lower resolution data to work with there, and it is heavily undersampled so needs blur to cover up what otherwise would show extreme aliasing.