The EPR Bell experiment protocol consists mainly in a comparison between the quantum prediction of a correlation function in \( c(\Delta) = Cos^2(\Delta)\) leading to entanglement and the so called triangle curve, defined as \( c(\Delta)=1-{{2 |\Delta|} \over \pi } \), assumed by Bell to be the best that a classical analysis might produce. Alain Aspect proved experimentally the quantum prediction.

But many physicists disagree with the Bell - Aspect experiment, arguing that this doesn't prove the inability of Classical Mechanics to predict the same in lab conditions. The burden of proof being on the defenders of the quantum entanglement, this objection is seriously considered by the experimenters.

There are more than 20 supposed 'loopholes', most of them adding other assumptions like observers quantum interactions and conscienciousness weird theories. Among these *loopholes*, almost all ruled out by hundreds of articles, the fair sampling ( AKA detection loophole ) persisted because it seems to be the less incredible. The defenders of this weakness argue that undetections *might conspire* to construct the desired correlations if the pairs cross detection rate is globally less than 75%. Others say \( {2 \over 3} \) and a few claim \( 2^{-{1 \over 2}} \).

Indeed, the toy models authors tend to idealize the contexts, interactions and measures. It was seen again in the recent debate around the Wigner friend paradox.

To close this option, experimenters must show that a detection rate greater than 75% doesn't affect the quantum prediction. This paper is intented to prove that it is possible to get both the rate and the quantum prediction.

Reading the report :

Closure of the fair-sampling loophole does not rely on space-time considerations and can be observed in the experimental data.

and :

Here we report a Bell test that closes the most significant of these loopholes simultaneously. Using a well-optimized source of entangled photons, rapid setting generation, and highly efficient superconducting detectors, we observe a violation of a Bell inequality with high statistical significance.

But in the supplemental material located here : we find these measures:

The angles of the polarizers are

\(a1 = 94°4\) , \(a2 = 62°4\) , \(b1 = -6°5\) , \(b2 = 25°5\).

The number of valid trials is \(N = 3,502,784,150\)

and the relevant counts are

\(N_{11} = N_{100°9} = 875,683,790; N_{11}^{++} = N_{100°9}^{++} = 141,439\)

\(N_{12} = N_{68°9} = 875,518,074; N_{12}^{+0} = N_{68°9}^{+0} = 67,941\)

\(N_{21} = N_{68°9} = 875,882,007; N_{21}^{0+} = N_{68°9}^{0+} = 58,742\)

\(N_{22} = N_{38°9} = 875,700,279; N_{22}^{++} = N_{38°9}^{++} = 8,392\)

We can see here that the detection level is less than 75% and also less than 1%. These values are not compatible with the conclusion. Not only the detection loophole is not closed but also the conclusions are false.

This experiment must be done again with a better detection rate goal. Perhaps it is unreachable because 75% of cross detection corresponds to 86% of detection by arm, which is a very high value with the current technologies. In this case, at least, the independence of the correlations and the detection rate must be shown.

I thank the authors for the complete raw data that was kindly provided on request.