Accuracy. If you could somehow prove that every vote could accurately be tied to a human pulling a lever, confidence in election results would be higher. Also instantaneous results and no issues of a recount (like in the 2000 election, where the refusal of a recount in a state cost Al Gore the election).
On the other hand, the way votes were counted back in the days of the paper ballot were not something to be proud of either.
Anyone who's been involved in counting votes has seen more vote count fraud than they could possibly try to explain to others. Everybody used to be involved in that, from the people counting who didn't want to be there and would do anything just to get over with it, to the party delegates, to the people in charge of the sections.
I understand all the criticism and I praise your work in pushing for a more secure and auditable stack but it's hard to argue that the previous system was better in any way.
I don't really see how that's superior to anything. It adds complexity to a system that's already hard to understand to large slices of the voters, and provides nearly no actual advantage to either the old paper ballot or the new electronic system.
A potentially better approach would be to have the systems themselves publicly auditable and somehow have the live ballot devices verifiable.
VVPATs do not add complexity, they cheaply allow a layman to verify if a proper record of his/her vote was produced.
I can't parse your last sentence, sorry.
PS: Yes, I would prefer to redesign the whole thing from scratch and tightly integrate physical and electronic records, as in an optical scanner, but this is very unlikely to happen anytime soon.
If they can ever be made secure, it's what I wanted to know from the researchers.