No matter what we do to these models, we can't define the boundary conditions very well for the purpose of testing.
WUWT has demonstrated pretty clearly that trying to make your model match the ground station data would be meaningless. Then you can resort to the balloon data and the satellite data, but its history is rather short.
Taking measurements over and over again using tools that measure (sextants, thermometers, rulers, tape measures, etc) allows you use statistics. Using a computer model to recalculate what something is and applying statistics to it is completely meaningless.
The EU is opening a system next year that will let you see what the weather will be like within a 5km area by the hour 50 years and more from today (post 2060). Take that same system and make it predict next month and start finding out how it is wrong and you have the start of a scientific process. Maybe not a perfect process, but closer than making prediction 50 years from now when we can't predict next year.
There would be tremendous value for a system that 2 months ago would have told planners that Interstate 5 would be covered with water sometime this month. As it was they sounded the evacuation order maybe a day before the water got dangerous. The problem with selling such a system is easy, IT WON'T actually predict anything. If the predictions are always wrong, NO ONE IS GOING TO USE IT.
A prediction 50 years from now? I won't likely be alive or remember the BS that happened now.