The idea behind the following picture is this:
Alice picks axis \vec{a} and Bob picks axis \vec{b}. You can always choose a coordinate system such that \vec{a} is in the x-direction and \vec{b} is at an angle \theta away from \vec{a} in the x-y plane.
The intrinsic spin vector, \vec{\lambda}, can be in any direction, but we can write it as \vec{\lambda} = \vec{\lambda_z} + \vec{\lambda_{xy}}, where \lambda_z is the component of \lambda in the z-direction, and \vec{\lambda_{xy}} is the component in the x-y plane. For the purposes of determining whether Alice and Bob get spin-up or spin-down, only \lambda_{xy} is relevant, so in the diagram, \lambda just refers to this component in the x-y plane.
So if \lambda is in the x-y plane, we assume that it has equal likelihood of pointing anywhere in the x-y plane.
So let A be Alice's result and let B be Bob's result. The first picture shows how Alice's result depends on \lambda: If \lambda lies anywhere in the yellow region, then Alice gets +1. Otherwise, she gets -1. The second picture shows how Bob's result depends on \lambda: If \lambda is in the red region, Bob gets +1, and otherwise, he gets -1.
The third picture shows the joint probabilities: Alice and Bob get the same result if \lambda is in the orange and white regions, which occur with probability \theta/\pi. Alice and Bob get opposite results if \lambda is in the yellow or red regions, which occur with probability (1 - \theta/\pi). So the product A B is +1 with probability \theta/\pi and -1 with probability 1-\theta/\pi. So the expectation value of A B is (+1)(\theta/\pi) + (-1) (1-\theta/\pi) = -1 + 2\theta/\pi
View attachment 114570