Not sure I like these projections because I won't be here to check them.

Still a worry with error terms. Maybe I have misread the methodology of others, but the "noise" you get from sampling once a minute all day long is bigger than the noise you get from reading a thermometer daily, which is bigger than the noise you get by taking proxies back 1000 years. But error is not the same as noise. If you state that 1998 was the hottest year for a 1000 years, would you not need to have yearly records (in a way that proxies do not seem to reflect) going back that 1000 years with a reading evey year? I mean, if people smooth enough with one of their 43 flavours of ice cream, you can end up with a straight line that might or might not have a slope.

As for correlation corfficients, I see words written like "a strong correlation of 0.51" - that type of thing. People get to use rules of thumb and it has to be 0.8 or better before I get too excited. Below 0.5 I just lose interest. Do you have a mental cut-off or would that be judgemental like I am?