The grail of realtime rendering is temporal coherence.
The fact that the majority of a typical video signal is similar between frames.
Search for the rendering grail ends thus in something which minimizes bandwidth (power) by maximizing usage of temporal coherence.
Which implies a temporally coherent scene structure,
such as a virtual shade cache mapped to a hierarchical scene graph (LOD) where each node is a "patch" of the scene.
Given visibility is also mostly temporally coherent,
implies a desire for cached "patch" to "patch" visibility
for fast light transport (GI).
Given a hierarchical scene graph,
cached visibility could be compressed,
for example marking nodes in at various levels as either all visible or all invisible (cuts or includes all children of node).
Which reduces light transport to a weighted gather operation with feedback in the virtual shade cache (multi-bounce GI).
Important part is the lack of searching.
The frequency at which visibility is solved, could be adaptive,
meaning improved across frames.
Shadows for things like moving folliage are often spatialy coherent,
as in the spatial neighborhood of likely blockers would typically be near in the hierarchy of the scene graph to prior frame.