While clearly many of the improvements result from signatures segregation, do all of them depend on it? Or are there some, that are implemented independently, that don't need signatures segregation? If all improvements depend on signatures segregation, then there's actually little sense in distinguishing the two meanings one from another.
Edit: In other words, can some of the improvements be implemented as a separate softfork, or maybe even without softforking at all?
My personal opinion: segregated witnesses add a new kind of transactions, which look like anyone-can-spend for non-upgraded node making it a soft-fork. So it is possible to choose completely new set of rules while still making it a soft fork. You now have two options:
- use the old rules for the new transactions making it only segregated witnesses. Any further updates require either a hard fork or a new witness script version and then you need to support both intermediate and new version.
- or learn from the problems in the past and fix some small but nasty things that were on the hardfork wishlist for some time, like O(n) hashing and signing the input amount.
I prefer the second option.
The much discussed witness discount also serves two things: incentivice spending UTXOs by making signing cheaper and giving a block size increase with a soft-fork. And it also discourages using the additional space for crypto-graffiti spam, since the discounted witnesses might not be stored forever. You have a one-time chance to take all these advantages, so why not take it? You can't introduce them in a second step without a hard fork.
I see the political problems - giving a signal against doing any necessary hard fork, making it harder to increase the capacity later. But from a technical standpoint bundling these changes makes sense.