Comment by kapitza

The data quality in natural "experiments" like ice ages are as dubious as the GCMs, in the sense that you're simplifying very complex systems into single numbers. You'd never take this data set seriously next to the data you could get if you could mount a scratch planet and actually control the inputs.