EgoExOR is the first dataset to combine egocentric and exocentric views in the OR, with gaze, hand pose, video, and dense scene graphs across 84k frames, enabling holistic, human-centric surgical scene understanding.