We get folks looking at the statistical methods used and finding them badly designed and poorly used. “Broken” comes to mind. I took it on myself to look into GIStemp (as I am a computer programmer who knows FORTRAN and after saying for 6 months “someone ought to look at it” decided “I am someone”.) What I found was a nightmare of crummy code and questionable methods. A complete lack of any kind of ‘test suite’ or ‘benchmark testing’ code. A level of amateurish code and testing methodology that would have caused me to stop the product from shipping in my shops. (I have managed software production commercially including software that got 4 patents and was used in production.) Eventually that lead to examination of the GHCN data set directly where even worse issues were found. chiefio at jo nova’s
Chiefio gets exactly why I have been a skeptic for years. Go read the whole thing.