The Go fuzzer, when it finds a failure, will also walk back and try to shrink inputs (and still trigger the same failure). Not sure how effective it is compare

The Go fuzzer, when it finds a failure, will also walk back and try to shrink inputs (and still trigger the same failure). Not sure how effective it is compared to other options, but I'm not totally missing out there. It also builds a corpus of interesting inputs over time (those that cause new branches to be taken, since that's its goal: improve coverage). I'm less sure about combinators.