Sure, it's a bit weird but not that weird as they have a huge codebase in a specific domain. Also, they still end up using a lot of Haskell code (and semantics) although they write in a thin layer of "DSL" on top of Haskell and GHC (which itself is a lot of "Haskell in production").
> Mu has a strict runtime, which makes program performance easier to analyse and predict – but prevents us from just copy-pasting some Haskell libraries that rely crucially on lazy evaluation.
They say their codebase includes 400k lines of Haskell code (plus 4.5M lines of Mu). Also, they use GHC which is a huge Haskell program, and they customize it with their own extensions etc.
As they copy Haskell library code and only sometimes modify it, I would count it as Haskell code until they modify it.
> Mu has a strict runtime, which makes program performance easier to analyse and predict – but prevents us from just copy-pasting some Haskell libraries that rely crucially on lazy evaluation.