Yes you're on the right track: there isn't any 'action' between the two systems, it's a correlation. However, here's the thing that fries my brain: for classical systems correlation means the pre-existing value of some property is correlated (e.g. if I take a random glove out of a pair, the moment I look at the one I picked I know the handedness of the other). In the quantum case there isn't a pre-existing value for the property you're measuring (because you're free to pick the measurement basis), and you can't say one measurement outcome caused the other because you can set up the measurements to happen outside of each other's light cone (so you are free to pick which measurement happens first by changing reference frame).
So: how does particle A "know" how to produce the correct correlation if 1) the value is produced 'on the spot' and 2) there isn't a well-defined causal order between measurement of A and B?
Eh, that's still just correlation across a state vector. Honestly, quantum entanglement is proof that the many worlds / simultaneous superposition interpretation of quantum mechanics is bollocks.
Really all QM says is that there's some probability distribution of an outcome, and you don't know what outcome occurred until you take a measurement.
For example, you could imagine a world simulator that uses probabilistic logic gates. Where with some probability a bit could be 0 or 1. Now you could say that the bit is in some superposition or is 0 in world A and 1 in world B, blah blah.
But that's dumb. Really, the bit is 0 or 1 with some probability of the universe simulator. And you don't know which, until you observe its value.
That's only half of the argument though. The other half is how does entanglement correctly coordinate the probabilistic outcomes, when there's no causal order? (meaning, you can't assume there's a sort of signal that travels between them, because in a different reference frame the signal would have to travel back in time)
No, it’s the whole argument. You could imagine a probabilistic gate splitter that outputs two bits. Such that bits AB are either 01 or 10 with a probability distribution. That is you don’t know which of the two states you’re in, but you do know that bit A is always opposite of bit B. You can then take the bits as far apart as you want. And then measuring bit A will tell you the state of bit B. Nothing collapsed across interstellar space. Bit A and B were always in a certain state. We just don’t know which until we take a measurement. And since they’re correlated, we only need to measure one of them.
Hehe I wish. You are forgetting that you can locally decide the measurement setting. For a setting the bits are (anti)correlated, for another one they are uncorrelated. If they always had a value, not only it wouldn’t work: even assigning a value per measurement setting wouldn’t work because it is what we call local realism, which is disproved by Bell’s theorem. Look up the GHZ game for an example of a system where you cannot assign pre-existing values consistently. (In the CHSH game, which is the equivalent of Bell’s theorem the explanation is a bit more subtle). https://en.m.wikipedia.org/wiki/Quantum_pseudo-telepathy
So: how does particle A "know" how to produce the correct correlation if 1) the value is produced 'on the spot' and 2) there isn't a well-defined causal order between measurement of A and B?