One issue with Greg Maxwell's argument is highlighted, which suggests that Graftroot delegation is not "strictly less powerful" than using a normal transaction. This is because Graftroot enables delegation in a way that cannot be fixed in the chain. By defining the semantics such that () is allowed, calling g-sig a "Graftroot transaction", and giving it transaction semantics, the proposal can become more optimized while still maintaining privacy. It is suggested that g-sig only needs to be a full transaction if it's used standalone as in (), but if we want to have g-script --g-sig--> script1 ---tx2---> script2 (which should be the common case) then just the bare signature and script1 would suffice, as in the original Graftroot proposal.