Every cosmological parameter extracted from the small-scale CMB passes through a subtraction: the unresolved extragalactic point sources, radio galaxies and dusty star-forming galaxies below the detection threshold, whose collective power contaminates the damping tail where the primary anisotropies fade. The subtraction is a model: source counts measured above threshold are extrapolated below it, with Poisson and clustered components fit jointly alongside the cosmology. The problem is that the foreground model and the cosmology trade against each other: analyses of Planck, SPT, and ACT spectra have repeatedly found that shifting the point-source treatment shifts the inferred damping-tail parameters, and inter-experiment comparisons reveal residual source power treated differently between pipelines, contributing to the documented parameter offsets between Planck's high-multipole likelihoods and the ground-based experiments.
The direction of risk is an overestimate built on an undercount: if the true faint-source population is more abundant than the extrapolated counts, the templates absorb genuine sky power incorrectly across frequencies, since the spectral assumptions used to project sources between bands are calibrated on the bright, detected minority. JWST's persistent discovery of unexpectedly abundant faint high-redshift populations, and the unresolved infrared and optical background excesses, both indicate the faint-end extrapolations in use undercount reality, propagating a quiet bias into precisely the multipole range where lensing smoothing, N_eff, and n_s are measured.
The standing is an unglamorous systematic with cosmological reach: not a headline anomaly but a recognized soft spot, increasingly important as CMB-S4-era precision shrinks statistical errors below the foreground modeling uncertainties that the source-count extrapolations carry.