Prior bias in the marginal posterior P_ϵ(ϵ | I₁,...,I_n) as function of S/N ν for different galaxy sizes r_h (in arcsec). The posterior assumes a uniform prior. Shown is the error δϵ = | ϵ − ϵ_true | of the inferred ϵ for a true ϵ_true = 0.3 as obtained by combining the marginal posteriors of 5 × 10³ exposures of the same galaxy with random centroid positions. The pixel size 0.1 arcsec equals the PSF size (Moffat). Galaxy profiles and GLAM templates have a Sérsic profile with n = 2: there is no underfitting.