While preparing some additional documentation for Iavor's custom type errors work (which has been merged; thanks Iavor!) I noticed that Dominique Devriese has raised some additional questions on the proposal [1]. In particular, Dominique suggests that perhaps TypeError should simply be of kind `ErrorMessage -> Constraint`. My understanding of the proposal is that the intention is that `TypeError`s should be usable on the RHS of type functions, like `error` on the term level. However, is this strictly necessary? Is there any place where you couldn't just as easily express the `TypeError` as a constraint? If not, it seems like this may be substantially simpler approach and totally side-steps the detection of `TypeError` in inappropriate places on the RHS. Regardless, it seems like this (and the other questions) is worth addressing in the proposal. Cheers, - Ben [1] https://ghc.haskell.org/trac/ghc/wiki/Proposal/CustomTypeErrors#Somedesignqu...: