I wanted to remain as close to the paper (which was written for Gofer) as possible and so I didn't define an instance of Monad, but did my own functions ((|-) for bind and result for return)

The grammar that the parser recognizes is something like this:

A -> charA

A -> number<A>A

A -> _

Some examples, and how they'd be parsed:

abc3<def> -> abcdefdefdef

pq5<a2<b>>op -> pqabbabbabbabbabbop

3<a4<b5<c>>> -> abbbbcccccabbbbcccccabbbbccccc

The input string is parsed into a Compressed value, which is then passed through the transform function. It looks for repeat blocks that are nested in each other without any text before or after them, and fuses them together.

...3<ab4<5<pq>>>... ==> ...3<ab20<pq>>...