Ah, friction, dissipation, entropy production. Life in the non-equilibrium world is so unpredictable…
Your comment: ‘Continuously readjusting non-explanatory models to fit new data isn’t what a science does. ‘ is too optimistic.
I guess by ‘causal mechanical’ you mean dynamical models, and explicitly not regression or stochastic schemes? If so, fitted models like the ones I mentioned are still the best tools available in some noisy physical sciences, I’m thinking hydrology in particular, and older but still used atmospheric subgrid parameterisations.
Nevertheless, you’re right to prefer a dynamical approach where available. I suspect though, as others in the thread have noted, that the system is simply unpredictable at some of the time scales we’d like it to be, and that it might turn out that simple models explain most of the predictable variance.