When does eventual consistency really bite you in a real-time system?

We always talk about eventual consistency, you know, as a trade-off for scalability in distributed systems, especially when you're trying to get real-time performance. But if we're being honest, in a truly real-time situation, where how happy users are really depends on them seeing the newest data almost instantly, when does 'eventual' just not cut it anymore? I'm picturing things like financial trading platforms or those tools where everyone edits a document together. At what point does the delay from eventual consistency, even if it's just a few milliseconds, cause more headaches than it fixes? And what do you even do then to fix those issues without giving up too much scalability? Is it just about read-after-write consistency, or is there something else at play? Let's talk about where that line actually is.