Like in that article John Baez had mentioned on G+ about the temperatures they use all sort of wild averaging, estimation and simulation methods, from part 4. Construction of stacks:

>We constructed the temperature stack using several different weighting schemes to test the
sensitivity of the temperature reconstruction to spatial biases in the dataset. These include an
arithmetic mean of the datasets (Standard method),
both an area-weighted 5°x5°and 30°x30°lat-lon gridded average, a 10° latitudinal
area-weighted mean, and a calculation of 1000 jackknifed stacks
that randomly exclude 30% and 50% of the records in each realization

and Fig 8 has a

>Simulated global mean temperature for the last 11000
years at the 73 proxy sites (black) from the
ECBilt-CLIO transient simulations (81)

where apparently something called accelerated transient boundary conditions was used.

The different methods are reflected in quite different curves (see e.g. Fig S25 page 34), where the differences seem to exceed
the temperature ranges as mentioned in the discussions above.