This is an interesting post. I was looking at the North America h5 maps on the gfs earlier and something really stood out. At that range (view) when you toggled back 4,5,6 runs you couldn’t tell any real difference. Made me realize that it comes down to very small features when you’re looking at the practical weather at any specific time. So I can see a scenario where a model may score well on a large scale evaluation but poorly on the small scale that actually determines the real weather you get.