Hand-counting.

It is _not_ about mistakes.
Mistakes in the manual counting don't matter for the outcome. One more or less vote for one party. Who cares. And well, even in case, you can recount.

What _does_ matter is the possibility to setup a _giant_ fraud based on a modified voting machine hardware/software. That fraud is _extremely_ difficult to detect, because in practice no voting machine operator does understand these machines. They blindly trust them. The CCC proved that several times here in Germany.

So the real difference between traditional voting and electronic voting is:
Traditional voting is extremely difficult to manipulate (so that it changes the outcome in practice).
But voting machines are (compared to traditional voting) extremely easy to manipulate. Even if you have checksums and seals. They often are simply not checked in practice by the operators.
And even a checksum can be manipulated.

So let's see, what does a checksum do? Basically nothing. It's printed on the screen or maybe on a sheet of paper. So well, if you have control over the whole machine (and you often do, as they are not stored securely. CCC showed that), you can print whatever you like on the screen or the printer.

The problem is not that you can make machines secure by doing cryptographic stuff and so on. The problem is that the human operators do not understand the system and do not operate it in a secure way.
And in my opinion you can _not_ change that. No way.

Or well, let's take another example. Here in DE you can request recounts. How does that look like traditionally? Well, the sheets of paper have to be recounted. You can _watch_ the people doing that. You can _check_ them for correctness.
How does it work with a voting machine? The operator presses the "recount" button and another sheet of paper is printed out. That is _not_ transparent and I have _no_ chance to check the result. I'm forced to _trust_ the machine. But I don't. I mean, I don't even trust my own code... :)