EDIT: irrelevant bits removed because Jess beat me to it.
In terms of mistakes, it's much easier to just leave them in and edit them out in post-processing than it is to stop a recording, back up, and start again. So just ignore lines with mistakes as long as there's a good take somewhere in there.
Thought-speech will be done by timers, and I need to go find the effect. It takes some work to do well.