Thanks for your code again!
The first action of a new episode is selected randomly in your Sarsa algorithm(Line 14 action=randi([1 2],1,1);), but referring to Sutton, R. S.'s book we should "Choose A from S using policy derived from Q(e.g.,epsilon-greedy)".From my perspective,the Q mentioned here is the updated Q of last episode,not the Q at the very beginning(if it is the Q at the very beginning, the epsilon-greedy equals to random policy). Are you trying to make more exploitation with random selected action in the first step or we might do as the book as well ?

Excuse me! What does Q-V Learning mean? The algorithm of Q_V_Garbage.m is more like the combination of TD(0) for evaluating v_pi and Sarsa control methods rather than the Q-learning method. Can you help to explain it?