The most important mental model to develop is a model of what can and cannot be feasibly optimized. Sometimes there really is no alternative to stuff like Rc<RefCell<>>, Arc<RwLock<>> or Arc<Mutex<>> - because you'll be working with pieces of data that have to be mutated from multiple places, and don't even have a well-defined lifetime so have to be reference counted. In that case, you should write a comment to that effect and move on. It's still good to see that RefCell<> and the like, because shared mutability is known to be a source of logical bugs. It's a bit like a low-key `unsafe`.