Very likely the issue is with the test infrastructure and not with the code under test.

My own fuzzing doesn't report any inconsistencies. But fuzzing is always necessarily incomplete.