Indeed. But I wholly expect that sometime in the next few decades machine learning will be able to identify real people and their online personas based on text patterns and such and not the current method that's based on leakage, or having a handle like "johnasmith" or something.

What you think is discrete today, may not be tomorrow.

Even today I'm sure Google and Facebook could identify you just based on traffic patterns (e.g., "chii" has lots of traffic on hacker news and then from the same location "john a smith" has traffic on facebook). They don't sell that info today, but perhaps in the future.