I don’t understand. If the audience member makes a change, and then the speakers events are played back on top of the change, the code won’t make sense.

If you make a change, then the screencast pauses and you create a branch of the codebase, so that you can experiment on your own.

Once you hit "play" again, your changes are reverted and you continue watching the teacher's code.

So the teacher's voice is never on top of your code, as that wouldn't make sense.

Did you try it?