Empirical distribution of the AIC-based relative model odds between the C_ℓ-fiducial and the C_ℓ-moments model from 100 simulations. We compare five different Galactic foreground templates, including the PYSM foreground model d10s5. Negative values indicate preference for the moments model. We find strong preference for the C_ℓ-moments model in the d10s5 foreground scenario, and only then.