SimSIMS is a first, brief, and impactful application of the neural classification-based Bayesian model selection technique presented earlier to pressing questions regarding host-dependent SN Ia standardisation and dust extinction. Concretely, it addresses the interplay between the possibility of an offset between the intrinsic brightnesses of SNæ Ia hosted by low- and high-(stellar-)mass galaxies (i.e. a mass/magnitude step) and different population models for the host dust, which may be influenced by other galaxy properties and give rise to an apparent (empirical) correlation with stellar mass. Using essentially the same simulator and neural network as in SIDE-real, we perform principled model comparison by deriving explicit posterior model probabilities (and hence, Bayes factors), having implicitly marginalised over 4000 nuisance parameters. Owing to amortisation, we explore the dependence of the results on underlying parameter values, thus visualising and quantifying Occam’s razor. When applied to the real CSP light curves, principled model selection prefers a model with a single dust law and no magnitude step, disfavouring (based on SN Ia data alone) different dust laws for low- and high-mass hosts with odds in excess of 100:1.