We're talking about predicting an effect which is 10^-5 orders of magnitude different from current configuration, with tools that were tested and failed within much shorter timeframes (~ months instead of years) and all of this is the only "evidence". The prediction abilities of these tools are only around the general trend, they failed predicting the magnitude correctly. So around 10^1 orders of magnitudes of accuracy.
All of this by computer models which have horrible code, all copying from each other, and are extended way beyond their empirical testing - while being fixed with fudge factors called "flux adjustments".
All of this in an extremely chaotic system. The kind of chaos which has no hope for even short duration prediction.
It's not science. It's a religion. There are no confidence intervals on this because if they had done a systematic analysis of the entire thing, they would have to admit that the chaotic behavior dominates their analysis by several orders of magnitude.
They do have plenty of data for what is actually going on - but the conjecture that it is CO2 causing it (as opposed to being just chaotic behavior, other factors, anything else really) has very little evidence. They abuse conjunction fallacy to convince everyone.