If a pathological workload can look completely reasonable and innocuous, then what's the difference between a bad-faith benchmark and a user making an honest mi

If a pathological workload can look completely reasonable and innocuous, then what's the difference between a bad-faith benchmark and a user making an honest mistake?