This sounds more like an argument for systems design. You can easily build models and meta-models that help navigate various scopes. For example the concept of the long tail is by definition a widely-scoped way of gaining leverage in a situation where smaller details are important.

Traffic control systems, computer hardware design...these all incorporate the same combinations of depth and breadth of thinking. I think it's wise to keep an open mind and ask whether it's really necessary to exclude a given model or entire set of them.

It’s not a question of excluding the model, it’s a question of the feedback loop where the model is used to both inform your decisions and validate they where correct.

Goodhart's law is an adage often stated as "When a measure becomes a target, it ceases to be a good measure"

> it’s a question of the feedback loop where the model is used to both inform your decisions and validate they where correct.

Who exactly are we criticizing for doing this? Seems a bit like a straw man argument tbh. I'm not even sure how that relates to the more general concept raised above, of a single model of reality being reliable.

If you want just one example, the IMF uses GDP data for critical decisions and to measure the results of those decisions.

At any granularity, knowledge consists of models in the consciousness of the thinker.

Direct observation exists alongside models. You may not be able to correctly interpret that observation, but it is still a form of knowledge.

This is true even if what’s being observed is by a simulated entity, the observation exists outside of it’s internal models and is indirectly related to things in the underlying reality. Ie, physical states of various bits of physical RAM.