In running the spherical renderer on a real dataset, I had to do some massaging to get a decent grid decomposition. That would have been a fair bit easier if I could instead have just supplied a list of grid objects (or maybe a list of covering or arbitrary grid objects?). I think there might be some value in adding a method for loading non-overlapping level 0 grids without the block traversal kicking in? This would come with the danger that a user supply grids that are too large to load... but that was happening to me with the default behavior when I tried to load a dataset that has a single large grid. While it'd be worth improving how yt decomposes a single large level 0 grid, I think yt_idv would also benefit from having a way to load an arbitrary list of grid objects.