In a layer-based editor, a correction applies to a stack, and its position in that stack is the only thing controlling what it sees. In a graph, the correction sees exactly what is wired into it. That difference is what makes it possible to grade a foreground element against the background it will be composited over, rather than grading both and hoping.
It also means a grade can be reused. Because a node's output is an ordinary frame, one corrector can feed several downstream branches, which is how you keep two elements matched without duplicating the settings that matched them.
And because effects accept a mask input, a correction can be limited to part of the frame without a separate node — including a mask produced by a keyer or a roto shape.