It does not explain anything physical - or there would not be dozens of threads arguing this.No. There is a single sample space for each pair of Bell alignments, but not all of them together. All can be carried out. @Morbert is referring to the fact that their refinement cannot be carried out.
It is directly relevant, per Fine's theorem it explains the violations of the Bell inequalities by QM.
When the data from an epr experiment are divided into the four categories each contingency table has expected marginals of 1/2 which is experimentally verified.