If you drop a coin it's affected by the air drag, bounces and tumbles on floor before it settles and you can read whether it was heads and tails.
If I understood it right, the Bell's theorem says that if the value what we'll read from a pair of coins is determined from the start, then the correlations between the two data will have limits given by the Bell's and other inequalities.
But in the coin tossing example there are lot of parameters that determine the result: air flows, the location it first hits the floor, etc. And these parameters can change in the meantime. So it's practically unpredictable to say at the time of tossing that it will be heads or tails. Pretty much similar situation what say about the particles. It's in a magical superposition state which then collapses and we can read heads and tails.
That's a bit similar to what we do in those particle experiment, we toss particles and measure whether they pass the polarizer or not or they'll be spin up or spin down.
So could we say the same about the experiments that aimed to show entanglement? That photon entering the polarizing beam splitter is passed back and forth millions times among rapidly thermally moving atoms of the crystal, which is quite chaotic. But at the end there is some order in the chaos since the two polarizations exit in two directions.
Can this chaos in that crystal cause correlations that violate the Bell's theorem?