I think you need slightly different terminology: given an ordered set of transactions T = t_1, t_2, .., t_n then Chunk(T) splits up that list, eg C_1 = t_1, t_2; C_2 = t_3; C_3 = t_4, .., t_n while maintaining its order (T = \sum Chunk(T)), and gives you a valid diagram (s(C_1) \ge s(C_2) \ge s(C_3)). Then if chunking a given ordering of txs results in a single chunk, ie Chunk(T) = C, then any reordering of T will result in a comparable diagram that is equal to or better than the diagram for T.
(That’s then obviously true because the diagram has the same start/end points, the first diagram for T is just a line, and the chunking inequality ensures the diagrams are concave)