It's interesting to know about what state machines you talk. From my experience most of the time it's an entity with state property with finger countable cardinality. And state is assumed to be changed directly. And it's not easy to reason because author only heard about state machines and state transitions are spread over all code base.
I was talking in the most general sense. I am sure there are state machine implementations that are terrible to reason about, especially any that emerge from a codegen tool. But hopefully they are the exception and not the rule.
Woah, this caught my eye:
> especially any that emerge from a codegen tool