Consider the following program: ```hs module T26277 where import Data.Kind type FunLike :: forall {k}. (k -> k -> Type) -> Constraint class FunLike p where myId :: p a a instance FunLike (->) where myId = id test x = myId @(->) x ``` With GHC 9.10, GHC 9.12 and HEAD, I get: ``` panic! (the 'impossible' happened) GHC version 9.15.20250723: typeOrConstraint a_aBk[tau:1] :: k_aBf[tau:1] Call stack: CallStack (from HasCallStack): pprPanic, called at compiler\GHC\Core\Type.hs:2743:14 in ghc-9.15-inplace:GHC.Core.Type typeTypeOrConstraint, called at compiler\GHC\Core\TyCo\Rep.hs:773:18 in ghc-9.15-inplace:GHC.Core.TyCo.Rep mkScaledFunTys, called at compiler\GHC\Tc\Utils\Unify.hs:802:33 in ghc-9.15-inplace:GHC.Tc.Utils.Unify ``` The problem is that we have a filled metavariable `k_aBf := TYPE kappa`, but several functions in the compiler expect the `TYPE` part to be directly visible (instead of requiring zonking); in this case it's the call to `mkScaledFunTys` in `matchActualFunTy`. If we change `matchActualFunTy` to do a full representation-polymorphism check (as in !14357), then instead we get an error like: ``` <no location info>: error: panic! (the 'impossible' happened) GHC version 9.15.20250801: unifyConcrete_kind: kind is not of the form 'TYPE rep' or 'CONSTRAINT rep' k_aBf[tau:1] :: * Call stack: CallStack (from HasCallStack): pprPanic, called at compiler\GHC\Tc\Utils\Concrete.hs:784:5 in ghc-9.15-inplace:GHC.Tc.Utils.Concrete unifyConcrete_kind, called at compiler\GHC\Tc\Utils\Concrete.hs:688:40 in ghc-9.15-inplace:GHC.Tc.Utils.Concrete hasFixedRuntimeRep, called at compiler\GHC\Tc\Utils\Unify.hs:179:32 in ghc-9.15-inplace:GHC.Tc.Utils.Unify ``` where again we have `k_aBF := TYPE kappa` but this is only apparent after zonking. Several functions in the compiler expect that the input type is of the form `TYPE rep` or `CONSTRAINT rep`, but I'm not sure how we are supposed to be upholding that.