You’re correct about one broken machine.
But code can erase itself and it can erase all evidence of itself from a drive by overwriting the sectors numerous times. Malicious code can coexist with, and interfere with, legitimate code. You can, for example, have separate code that accesses the same database. When it deletes itself, the legitimate code is still there. You could have code that changes one or more stored procedures in the database then, at the end of the night, changes them back and deletes itself.
All of that of course depends on the details of particular machines. Do they use a hard drive or flash ram? If they use a hard drive is it a physical drive or an SSD?
I don’t need an education about binaries, executables, etc. I’m a software developer.
I also don’t think it’s likely that the machines were hacked, it would be risky as hell because, as you say, one broken machine could give the whole thing away, and there would have to be a lot of hacked machines (of course someone would have to bother to check them). But I don’t think that it’s totally impossible, especially for a sophisticated nation state like, for example, Russia. In fact, they might not mind so much if it was discovered. Imagine the chaos if a whole election was invalidated. Putin would love that.
Part of a forensic audit would be to look at machines that broke or were shut down early for some reason, or weren’t brought into service.
My other concern is with bugs. I can tell you from experience that all the testing in the world does not guarantee bug free code. However, it seems unlikely that there would be some unintentional bug that would always favor one party over another.
What it comes down to is that I’m uncomfortable without some sort of meaningful paper trail. You always talk about people having confidence in elections, well voting on a black box with proprietary code doesn’t exactly lend confidence.