I really liked how coupling is described as “knowing.” I find we talk about “does x need to know about y?” more than we do “is x overly coupled to y?” because the former is a relatable indicator of the latter.

I can relate to this, having developed a coupled socio-emissions-carbon-climate model, which evolved for 20 years in java, until recently converted to scala3. You can have a look here. The problem is that “coupling” in such models of complex systems is a ‘good’ thing, as there are feedbacks - for example atmospheric co2 drives climate warming but the latter also changes the carbon cycle, demography drives economic growth but the latter influences fertility and migration, etc… (some feedbacks are solved by extrapolating from the previous timestep - the delay is anyway realistic). There are also policy feedbacks - between top-down climate-stabilisation goals, and bottom up trends and national policies, the choice affects the logical calculation order. All this has to work fast within the browser (now scala.js - originally java applet), responding interactively to parameter adjustments, only recalculating curves which changed - getting all these interactions right is hard.
If restarting in scala3 I’d structure it differently, but having a lot of legacy science code known to work, it’s hard to pull it apart. Wish I’d known such principles at the beginning, but as it grew gradually, one doesn’t anticipate such complexity.