Let $X_1,X_2,\ldots,X_n$ be a time-homogeneous discrete-time ergodic Markov chain on a finite state space $\mathcal{S}.$ You can assume stationarity and time-reversibility as well, if you like.
Fix $s_0\in\mathcal{S}.$
Let $A=\{1\leq i\leq n-1: X_i = s_0\}.$
Conditioned on the event that $|A| = m,$ is it true that the random variables $\{X_{i+1}: i\in A\}$ are i.i.d. according to $P(X_2\in \cdot|X_1=i)$?
The answer is NO because, for instance if $m = n-1,$ then we know that $X_{i+1}$ for $i\in A$ is mostly equal to $s_0,$ except perhaps for when $i=n-1.$ $\{m = n-1\}$ may be a rare event, but under its occurrence, the statement is false.
Yet it appears that something like this must be morally true, intuitively speaking. Is there a quantitative version of this intuition? For instance, if $m$ is much smaller than $n-1,$ can we say that $\{X_{i+1}: i\in A\}$ are in some sense close to i.i.d. according to $P(X_2\in \cdot|X_1=i)$ or at least that their relative proportions are close to the multinomial distribution with probabilities $P(X_2\in \cdot|X_1=i)$?