Learning goal: Vary transitions to inspect stationarity and forgetting the start. Add hidden history to test when a two-state Markov description fails.