We investigate the effect of foreground residuals in the WMAP data (Bennet et al. 2004) by adding foreground contamination to Gaussian ensembles of CMB signal and noise maps. We evaluate a set of non-Gaussian estimators on the contaminated ensembles to determine with what accuracy any residual in the data can be constrained using higher order statistics. We apply the estimators to the raw and cleaned Q, V, and W band first year maps. The foreground subtraction method applied to clean the data in Bennet et al. (2004a) appears to have induced a correlation between the power spectra and normalized bispectra of the maps which is absent in Gaussian simulations. It also appears to increase the correlation between the dl=1 inter-l bispectrum of the cleaned maps and the foreground templates. In a number of cases the significance of the effect is above the 98% confidence level.

We propose two methods for obtaining the dual of non-linear relativity as previously formulated in momentum space. In the first we allow for the (dual) position space to acquire a non-linear representation of the Lorentz group independently of the chosen representation in momentum space. This requires a non-linear definition for the invariant contraction between momentum and position spaces. The second approach, instead, respects the linearity of the invariant contraction. This fully fixes the dual of momentum space and dictates a set of energy-dependent space-time Lorentz transformations. We discuss a variety of physical implications that would distinguish these two strategies. We also show how they point to two rather distinct formulations of theories of gravity with an invariant energy and/or length scale.

The COBE-DMR 4-year maps displayed a strong non-Gaussian signal in the ``inter-scale'' components of the bispectrum: their observed values did not display the scatter expected from Gaussian maps. We re-examine this and other suggested non-Gaussian features in the light of WMAP. We find that they all disappear. Given that it was proved that COBE-DMR high noise levels and documented systematics could at most {\it dilute} the observed non-Gaussian features, we conclude that this dataset must have contained non-negligible undocumented systematic errors. It turns out that the culprit is a combination of QuadCube pixelization and data collected during the ``eclipse season''.