Artificially created bugs used to evaluate whether repository guidance helps agents locate and fix issues correctly.