We've begun to encapsulate data manipulations by domain, and give adapters more flexibility to choose which to apply.
Perhaps we can move normalization functions into the transformers, and introduce a new interface. The benefit is that we put all code having to do with e.g. email address in one place. The drawback is that we don't need the flexibility yet, adapters seem to never need custom normalization logic, and on the other hand we might want to normalize using standard logic, but do staging using custom logic, which would require splitting the encapsulation anyway.