EM - I don't know about your example, but I understand that something like 11 out of 12 Met Office seasonal forecasts were completely wrong due to the use of overheating models.

They gave up publishing their forecasts after years of humiliation.

Schrödinger's cat

It was Martin A's example. I just checked the numbers.

I presume you know that seasonal forecasts are the most difficult. They are far enough ahead that variations in weather have time to build up, while too short for long term trends to become apparent.

A ten day or ten year forecast is much easier.

Simon Hopkinson.

The forecast Martin A referred to was more complex than a coin toss and the odds of getting that close by chance were. lower.

"Lord Lucan death certificate granted"

I hope we don't have to wait 42 years to see the death certificate issued for Trenberth's missing modeled heat.

A guest blog over at WUWT discusses the fact that Federal government spends about $1.5 billion on GCMs and is beginning to question the lack of documentation and validation that would normally accompany other scientific models.

The article makes this point:
"No typical scientist or engineer would apply significant trust to models which do not provide transparent calibration documentation. Many, including myself would argue further that even models which do show skill may merit some trust, but cannot be trusted very far. How then do the premier United Nations Intergovernmental Panel on Climate Change (UN IPCC) GCMs provide neither model accuracy nor transparency in association with their forecasts?"

I and others here have been making exactly the same points.