I am looking at Quantum Teleportation is a Universal Computational Primitive, and I can't follow how they derive the circuit for Fig 2.
I understand that the CNOT gate between the second qubit of the first bell pair and the first of the second occurrs after the bi-product operators acting on qubits 3 and 4. We can then use commutation to move it earlier in the circuit. However, taking a CNOT gate and moving it through $X$ and $Z$ bi-product operations, we get (going from left to right, but it doesn't matter)
$$C_{3,4}Z_{3}X_{3}\to Z_{3}X_{3}X_{4}C_{3,4} $$ given classical control by $|\alpha\rangle $ and $$C_{3,4}Z_{4}X_{4}\to Z_{3}Z_{4}X_{4}C_{3,4} $$ given classical control by $|\beta\rangle$
Even taking the control and target for the EPR pairs as flipped, we get
$$C_{4,3}Z_{3}X_{3}\to Z_{3}Z_{4}X_{3}C_{3,4} $$ given classical control by $|\alpha\rangle $ and $$C_{4,3}Z_{4}X_{4}\to X_{3}X_{4}Z_{4}C_{3,4} $$ given classical control by $|\beta\rangle$
But they have
$$C_{4,3}Z_{3}X_{3}\to Z_{3}Z_{4}X_{4}C_{3,4} $$ given classical control by $|\alpha\rangle $ and $$C_{3,4}Z_{4}X_{4}\to X_{3}X_{4}Z_{3}C_{3,4} $$ given classical control by $|\beta\rangle$
Now given these are all LOCC gates, I know that in certain instances, depending on the measurement outcomes, these would cancel to give the usual single $X$ and $Z$ operations. Additionally, they are bi-product operations, so it doesn't really matter what happens with them. But I would like to understand why I am getting something different than they are for the new LOCC operations with classical control.