Breaking the Supernova Dipole: Why Late-Time Anisotropy Claims Collapse Under Harmonic and Variance Decomposition

Recent analyses of Type Ia supernova samples have reported statistically significant dipole anisotropies in Hubble residuals and have interpreted them as evidence for cosmic bulk flows or anisotropic expansion. I reanalyze the Pantheon+SH0ES compilation using a harmonic decomposition framework that explicitly separates mean-field anisotropy, variance-field anisotropy, and higher-multipole leakage. I introduce a "narrow-path" dipole estimator that projects out l ≥ 2 modes and test its significance using adversarial nulls that preserve redshift structure while breaking sky–residual coupling. I find that the apparent dipole signal collapses once higher multipoles and anisotropic noise are controlled. The variance field exhibits dominant l ≥ 2 power, strongly correlated with survey geometry, and the dipole does not survive redshift-stratified shuffle nulls. I conclude that the reported dipole is not a physical flow but a weighted average of inhomogeneous uncertainties, and that late-time supernova anisotropy claims do not resolve the Hubble tension.