If you ask Fable or 5.6 Sol to improve performance, it will generally know to build a benchmark and create a test corpus. I'm not sure where the contrary suggestion is coming from.