> As for coercion, there are really two types: coercing someone into voting a certain way, and coercing them afterward to reveal how they voted.

These amount to the same thing. The ability to reveal how one votes is what allows them to be coerced.

> It might suck if the coercer doesn’t believe you and you really did delete them, but at that point thugs beating you up is kind of its own separate problem.

The point of the secret ballot requirement is to make potential thugs aware that evidence of how someone voted does not exist. It is this problem that creates the need for the secret ballot requirement.

> If the coercion is about making you vote a particular way, some schemes let you vote multiple times and only count the last one, so you can just vote again after the coercer leaves.

You can do this in the poll booth as well; because the ballots cannot be distinguished, you can fill out a fake ballot and simply avoid depositing it. But being able to vote multiple times in an online election doesn't prevent coercion, because presumably the election has to have a result and voting can no longer occur after a certain time; so the coercer only needs to wait until after the election results are "locked" before requesting proof.

The first paper you link is a marked downgrade from existing voting systems which can achieve ballot secrecy without requiring fake ballots.

The second paper is more fleshed-out from a voting systems standpoint, but it makes a poignant qualification around registration security:

> Network-level anonymity remains vulnerable to timing correlation attacks. An adversary observing network patterns could potentially link registration and voting transactions from the same IP address. Mitigation strategies could include randomized transaction delays, Tor integration for network anonymity, or mixing services at the application layer. However, these additions would significantly complicate the user experience and mobile deployment.

And this is what makes electronic voting so hard to recommend even as a computer scientist. I think folks don't truly understand the breadth of side channels we create when we take a physical process (show ID, fill out ballot, drop in box) and move it to the Internet. The side channels are concerning because they can be observed and correlated without getting caught or impeding the process, unlike someone planting cameras or following you into the voting booth.

These do seem like things that could be solved eventually. But really I'd say my take is that you're not putting the same effort, and probably since you have more knowledge of the digital side, have a bias, towards seeing its flaws while not putting the same effort in seeing the flaws in the current process.

Trade offs, that's the known answer to all comp sci problems :p. Here too it applies.