Wow nice. If an llm could replicate these excellent examples, I'd consider the pelican benchmark fully saturated.