Markov Chains presume a stable / stationary distribution.

The recent Veritasium video on Markov Chains was great.

It made me realize that Markov Chains work great for stationary distributions.

That is, systems without an internal feedback loop.

Many systems have a compounding component.

Every system has some kind of feedback loop with its surroundings.

How well the "isolated system" assumption applies is how slow and weak the feedback loop with the environment is.