We present a comprehensive Bayesian test of statistical isotropy in CMB using end-toend posterior samples from the BeyondPlanck framework. Extending the power-tensor methodology, we replace single best-fit maps with full posterior ensembles, thereby propagating uncertainties from noise, gain fluctuations, and foreground separation directly into anisotropy diagnostics. We examine the quadrupole–octopole alignment, low-ℓ axial alignments, and high-ℓ power entropies, and we generalize the power-tensor formalism to E-mode polarization to provide the first probabilistic assessment of polarization isotropy from power-tensor statistics. While the BeyondPlanck maximum-a-posteriori solutions reproduce previously reported anomalies, in power-tensor alignment metrics, the marginalized posterior distributions of these statistics exhibit substantially broader credible regions. Within the power-tensor framework, this broadening reduces the apparent level of tension with ΛCDM once instrumental and astrophysical covariance is incorporated. A comparative study with Cosmoglobe and NPIPE demonstrates that several apparent anisotropic features correlate with specific systematic modes, particularly time-dependent gain variations. Our results suggest that high-significance anomalies found in traditional frequentist analyses may be overestimated due to incomplete treatment of correlated uncertainties, and that a Bayesian sample-based power-tensor analysis provides a more robust characterization of large-scale CMB isotropy.
Akash Gandhi (Wed,) studied this question.