Please contribute examples of fuzzing and shrinking bugs

[mgold]

It seems that the implementation of fuzzers and especially shrinkers is constantly in flux. But, we also appear to be designing blind.

I am asking for real-world examples of when 100-run fuzz tests were unsatisfactory. That is, SSCCEs of code and test where the tests either do not detect a bug or shrink to an unhelpfully large counterexample. Also indicate the (human-derived) smallest failing value that the fuzzer did not find.

Again, real-world examples: something that came up in production Elm code or caused you a significant amount of troublesome debugging.