The giveaway is this sentence:
"No longer are models judged by how well they reproduce data from the real world-increasingly, models provide the data."
This is wrong. One of the first things you do when you make a model is to dial it back to a previous date and see if it accurately predicts what was observed thereafter.
We saw that the Hurricane centre for example successfully predicted the track of every hurricane this year - to within a few miles and sometimes weeks ahead. The models are not only increasing in complexity, they are increasing in reliability. The amount of computer power is increasing at a phenomenal rate - so the models now are miles better than a few years back...a few months even...
What they are producing is consistent results from entirely different models..that's why the alarm bells are ringing.
Dave L