Sure. And for what you described, that works fine. But now, what if instead of measuring both oreos on the same axis, we measure then on slightly different axe

Sure. And for what you described, that works fine. But now, what if instead of measuring both oreos on the same axis, we measure then on slightly different axes? In particular, let C(a) be the correlation coefficient between the orientation of the oreo measured along the z axis and its orientation measured at an angle a to the z axis. You can derive a limit on the correlation that can be seen between any two angles (a and b, assume for simplicity coplanar with the z axis) - assuming the state of each oreo is a function of local variables like its speed and axis of rotation: |C(a) - C(b)| - C(b - a) <= 1 For example, if you pick a and b both = 0 (measure everything along the positibe z axis, like you do in your comment), then the left hand side is just |(-1) - (-1)| - (-1) = 1, i.e. the QM prediction for that is perfectly replicable by tasty, tasty oreos. But here's the thing: quantum mechanics breaks that limit for some angles. E.g. try it with a = pi/3, b = 2pi/3. Left hand side comes out as 1.5. Informally, the absolute best correlation between a and b you can get with oreos is perfect linear correlation - what the inequality above is = 1. But with QM, you get a sine wave! So rotating oreos seem like a good analogy to entanglement at first, but it's misleading. Turns out, neither rotating oreos nor any other classical system of local hidden variables can actually replicate the results of entanglement in QM. That's (a simplified form of) what Bell proved.