While there are existing fuzzers like javafuzzer that find bugs in C2, we find that the generated programs take quite a while to find and are overly contrived, whereas most C2 bugs (also those found by a fuzzer) have a rather small reproducer. Our hypothesis is that a fuzzer that instruments C2 for feedback (e.g. what optimizations took place) and takes existing test cases as seeds, could be guided to quicker and smaller finds. Further, such a fuzzer should also cover new java features currently not covered by other fuzzers we are employing.