The Choice Gradient Sampling algorithm shown in the paper could be used to integrate/exploit feedback (coverage, etc.).
Hypothesis can use a fuzzer as its input source, but it's marked as experimental.
It doesn't seem like "proper" fuzzers are used very often with Haskell (e.g. there's an example at https://www.tweag.io/blog/2023-06-15-ghc-libfuzzer/ ), though many people seem to have hooked up the HPC code coverage tool to property checkers, e.g.
