Agreed that devirtualization exists and improves performance, but it doesn’t address the semantic problem in that interfaces are about heterogeneity (and thus dynamic dispatch) while contracts are about homogeneity (which naturally lends itself to static dispatch). Additionally, devirtualization is a lot harder to implement (at least with any sort of useful comprehensiveness) as the compiler needs to prove to itself that the concrete type never varies which implies plumbing information across compilation units and (I think) breaking independent compilation of compilation units or at least making compilation dependent on some earlier whole-program-analysis pass.