Research questionHow can Gaussian splatting render unseen viewpoints reliably when large-scale scenes have incomplete, unstructured view coverage?Large-scale scenes are often captured from views that cover most content while leaving certain regions insufficiently observed. Representations built from these images may render inconsistently or implausibly from viewpoints that were not sampled.