The dust between us and the primordial sky has polarization habits nobody ordered. Planck's 353 GHz maps, the definitive census of polarized Galactic dust, revealed statistical regularities the interstellar-medium models did not predict and still do not reproduce cleanly: the dust's E-mode power exceeds its B-mode power by a factor of two (EE/BB near 2, where naive turbulence expected parity), the TE correlation is systematically positive, and, most dangerously for cosmology, the dust pattern decorrelates between frequencies, the polarized sky at 353 GHz is not a perfect template for the dust at 150 GHz, with decorrelation parameters that vary across sky regions and analyses. Every search for primordial B-modes lives or dies by extrapolating dust templates across frequency, so EE/BB structure and frequency decorrelation are not ISM curiosities: they are the systematic floor under the tensor-to-scalar program, the same floor BICEP2 fell through.
The standard-model reading treats these as astrophysics to be modeled: MHD turbulence simulations tuned with alignment physics can approach the EE/BB ratio, filament-based models reproduce pieces, and decorrelation is parameterized rather than derived. The persistent gap is structural understanding: the dust statistics encode the Galactic magnetic field's three-dimensional architecture, multiple misaligned field layers along each sightline produce both the EE/BB signature and frequency decorrelation, and the field architecture itself is an unexplained input. The B-mode program's precision now exceeds the foreground theory's.
The standing is a foreground problem promoted to a physics problem: LiteBIRD and CMB-S4 forecast r-sensitivities that decorrelation at the measured-to-allowed levels can fully counterfeit, and the dust's organized statistics remain, at bottom, a statement that the local universe's magnetic architecture is more structured than the models that try to subtract it.