Property based tests tell you the invariant held on the cases the generator thought to produce, which is the same blind spot as an example test with better coverage. Get the generator itself under scrutiny, because a shrinking strategy that only explores the shapes you encoded will happily report a pass for a domain it never entered. I've found the time saved on writing examples is better spent there than on starting the next feature. What does your generator know about the input space that the code under test does not?