Here is a technical description. I’m reading through it now, so perhaps comments later.
[Update a while later]
Sorry, link is fixed now.
[Update late morning]
OK, I skimmed it. As one would expect, there’s a trade off between development costs and ops cost with regard to lander reusability. Ultimately, to get to low marginal costs, we have to not throw hardware away. I also wonder how much it would take to make the Centaur reusable over a period of months or years. They’ve got a start on it with the refueling scenario. Eventually, if one is getting propellant from the moon, that would make sense. I would have liked to see a trade between LLO and EML-1 or EML-2, though. It looks to me like they settled on LLO early on.
I’m amused that they have to defend their costs as being “too low.” They look high to me (a hundred million for training?), but I have vastly different expectations about these things.