"You're reading the results wrong"

"Changing the traversal order destroys the same initial conditions"



PS: While messing around with the Python version, I also simply added a fresh shuffle of the blocks order, between the two methods. Guess what happened? The results were almost a perfect 50 - 50, just like expected. But I didn't mention this so to not get "not the same conditions". Your experiment, with using a totally random useless prefix to get the same biased results, is much much better though.