Comparison between log r_eff,90% and log r_eff,68% dust disk sizes for smooth disks (left panel), an entire set of substructured disks (middle panel), and substructured disks selected from our best case (i.e., 10^−3.5 ⩽ α ⩽ 10^−2.5, 10^−2.3 M_⋆ ⩽ M_disk ⩽ 10^−0.5 M_⋆, υ_frag ≥ 500 cm s⁻¹, m_p ≥ 150 M_⊕, and r_p ≤ 0.75 r_c) (right plot). Blue dots represent a subset of the disks used in the estimate of the correlation between r_eff,90% and r_eff,68%. The best fit obtained exploiting the linmix implementation of the Bayesian linear regression method developed by Kelly (2007) is shown as a black line, with a 1σ confidence interval reported in red.