[dagss]

This is mainly Gurzadyan's work, and it shows such a complete lack of understanding for basic statistical analysis it's not even funny. They basically assume that if you have an i.i.d. random vector x, then a linearly transformed vector z = Q x, with Q some linear map, should also have coeffients without correlations.

For a start look here: http://arxiv.org/abs/1105.5051 -- beyond explaining the fallacies it has references to 3 independent analyses that failed to reproduce the result on the same data set.

The only interesting discussion to be had here is one about journals and peer review practices (i.e., how A&A could let something like this slip through).

That said, the randomness (or nature of randomness) in the CMB is actively researched (such a how close the random signal is to Gaussian). And anything you think is "CMB" run the risk of being foregrounds or systematics which would not be "random" in one sense.