Monad transformers in Haskell from the ground up. Nothing imported, just code.
Current development is in directory haskell/nilsson. Tests are provided, using hunit.
The idea is to implement the algorithm described in [1] avoiding the official mtl stuff and using instead the Lego blocks described in [2].
Why's that? The mtl library features transformers which implement a lot of each other's interfaces. This allows to invoke operations with little care about which layer of the monad pile we're in.
I believe that, in order to understand this complicated matter, it's better to start by keeping the various layers clearly separated and use lift consciously whenever we need one or another operation.
[1] http://www.cs.virginia.edu/~wh5a/personal/Transformers.pdf
[2] http://www.cs.nott.ac.uk/~nhn/MGS2006/LectureNotes/lecture03-9up.pdf
[3] http://blog.sigfpe.com/2006/05/grok-haskell-monad-transformers.html