While it is clear how to compare theoretical predictions of a given model with observations, the reconstruction of a phenomenological model from the data is a more subtle issue. The basic problem is that in the CMB, we see the whole history of the evolution in redshift projected onto the two dimensional sky. The reconstruction of the evolutionary history of the universe might thus seem an ill-posed problem.
Fortunately, one needs only to assume the very basic properties of the cosmological model and the gravitational instability picture before useful information may be extracted. The simplest example is the combination of the amplitude of the temperature fluctuations, which reflect the conditions at horizon crossing, and large scale structure today. Another example is the acoustic peaks in the temperature which form a snapshot of conditions at last scattering on scales below the horizon at that time. In most models, the acoustic signature provides a wealth of information on cosmological parameters and structure formation ([Hu & White] 1996). Unfortunately, it does not directly tell us the behavior on the largest scales where important causal distinctions between models lie. Furthermore it may be absent in models with complex evolution on small scales such as cosmological defect models.
Here we shall consider how polarization information aids the reconstruction process by isolating the last scattering surface on large scales, and separating scalar, vector and tensor components. If and when these properties are determined, it will become possible to establish observationally the basic properties of the cosmological model such as the nature of the initial fluctuations, the mechanism of their generation, and the thermal history of the universe.