Nice job~ in my Show HN earlier today, somebody linked me to something similar: The Aaronson Oracle which also tests the player with a binary predictor.
That was enjoyable because I was able to out-trick the tool until it caught on and began to improve in performance. Perhaps humans aren't the best with randomness after all. But then again, we are limited to two choices here.
Yes, at first it's easy to think you can win, but then very quickly the model catches up and learns the user patterns, it's kinda wild realizing how predictable we can be