There are three types of fluctuations: scalars, vectors and tensors, and four observables: the temperature, E-mode, B-mode, and temperature cross polarization power spectra. The CMB thus provides sufficient information to separate these contributions, which in turn can tell us about the generation mechanism for fluctuations in the early universe (see §4.5).
Ignoring for the moment the question of foregrounds, to which we turn in §5.2, if the E-mode polarization greatly exceeds the B-mode then scalar fluctuations dominate the anisotropy. Conversely if the B-mode is greater than the E-mode, then vectors dominate. If tensors dominate, then the E and B are comparable (see Fig. 15). These statements are independent of the dynamics and underlying spectrum of the perturbations themselves.
The causal constraint on the generation of a quadrupole moment
(and hence the polarization) introduces further distinctions.
It tells us that the polarization peaks around the scale the horizon subtends
at last scattering. This is about a degree in a flat universe and scales with
the angular diameter distance to last scattering.
Geometric projection tells us that the low- 
tails of the polarization
can fall no faster than 
, 
and 
for scalars, vectors
and tensors (see §3.2).
The cross spectrum falls no more rapidly than for each.
We shall see below that these are the predicted slopes of an isocurvature
model.
Furthermore, causality sets the scale that separates the large and small angle temperature-polarization correlation pattern. Well above this scale, scalar and vector fluctuations should show anticorrelation (tangential around hot spots) whereas tensor perturbations should show correlations (radial around hot spots).
Of course one must use measurements at small enough angular scales that reionization is not a source of confusion and one must understand the contamination from foregrounds extremely well. The latter is especially true at large angles where the polarization amplitude decreases rapidly.
