We have no good benchmark to estimate the bugfixing ability, it was mostly zero-short "in this case it works" example.