In the description, I would specify that the first length is meant to be the longer one.

The random tests expect 'False' when the first number is smaller than the second one.

The huge pointscore definition is probably not the best way

duplicate issue below

This introduces an Ord a requirement that is not in the spec and is not necessary according to the spec.

I would not consider this a Best Practice and there should be a test against it.

smart. makes it so simple

There are multiple problems with the Haskell version, the solution is also available for you to directly access.

upper case letter do not need to be included, all letter are lower case

instead of twice you could use
liftA2 max snd (dofast . fst)
or
max <$> snd <*> dofast . fst

Good point, I guess for no reason.
I was probably using it for something else in which I needed to specify that a is Fractional, and then I just left it as it was :P

Why not use fromIntegral?

Thanks! It works now, indeed.

The Haskell reference solution was incorrect. It should be good now.

I got this in the random tests:

Falsifiable (after 41 tests):
expected: 2
but got: 1
"xehtzzbajpcze"
"cjjsshesortelnxefoqgdogybqurdiqtyo"

The character 'z' does not belong to the second string, and the first string has more then one.
So, couldn't P1 win by picking 'z', and then again another 'z' after whathever P2 does?

Fixed

The Haskell version is still unsubmittable.