CosmoCoffee

This paper appeared on astro-ph today, with some fairly aggressive claims about detection of compact topology. Essentially, they compute the excursion sets (that is, regions with T > T_0), and plot connected regions with more than, say, 200 pixels. At least, that's what I think they do.

Then they find that lots of connected regions start popping up on two opposing hemispheres, once they start increasing T_0. Their axis is found to be centered on l=92, b=30 -- which, unfortunately, is very close to the ecliptic pole (l = 96, b=30), which also happens to be the symmetry axis of WMAP's scanning strategy.

A more natural explanation than a closed universe (to me, at least), is therefore that they are starting to see the CMB through the instrumental noise as they adjust T_0. This happens in the ecliptic poles earlier than in the ecliptic plane, because the noise is lower there. So a first good test for this analysis, I think, is to do the same with WMAP-like simulations, taking into account the WMAP hit count pattern..

exactly...you simply shouldn't treat any real CMB observation as a Gaussian random field with uniform variance, if you do you'll find all kinds of anomalies. Claims like this have little meaning without a proper comparison to simulated observations particularly when you find a signal which is correlated with the scan strategy. It's the cosmological equivalent of reporting a lab measurement without stating any errors.