Bizzare long-term behavior?

[hdrake]

Out of curiosity I ran the model out to 10000 CE. The model has reasonable behavior until about 3000 CE but then the temperature anomaly becomes increasingly negative and eventually breaks the constraint on damages because T becomes too negative... I tried adding a constraint that T>0 at all times but didn't seem to work (despite the solution apparently finding an optimum?)...

I should add a branch with a notebook exhibiting this behavior at some point.