Yes, that's more the issue, though whar I'm also saying is that we may never get to observations that can be meaningfully fit to 3 decimal places in astronomy-- fits at that level may always have a "google science" flavor. By that I mean the model may fit the data, but only at the cost of specifying parameters at a conceptually meaningless level of precision-- given the missing physics that is inevitably involved.
I am not against elaborate models, any more than I'm against using a net to try and catch a great white shark. But once you've caught it in the net, there's a need to isolate it from everything else in the net, and learn about its attributes. Then the next time one ventures near, perhaps a better and more targeted type of net can be designed. But what I see more often is the attitude "we got the shark, everyone can go back in the water".

In your example, we may find that certain kinds of dust are responsible for certain features. We can know that from the kitchen sink simulation simply by noticing the difference with and without that type of dust. But is that really all we want to know? Is it enough to say that dust "is included", and leave it at that? Or do we want to know why that type of dust makes that feature, and are our assumptions about that process really reliable, just because they allow us to fit an observation with enough tweaking? That kind of investigation is carried out by people who are interested in that type of dust, if there are such "shark enthusiasts" in the community-- but they tend to be found in "niche markets" that don't connect with the AAS meeting plenary talk on how nicely the kitchen sink simulation fits the data. Now, I'm not saying all astronomers have to address all questions, but I am saying that just because a kitchen sink simulation can fit some data by "including grey dust", it might not mean a whole heck of a lot by itself.
I think the disconnect there is that you are still referring to a rather different type of "simulation" than I am. Observers do talk about "simulating an instrument" and "modeling their data to remove noise" and that sort of thing, and those simulations and models may need to be quite elaborate. I'm talking about a simulation of a physical system that you are trying to understand by comparing the ramifications of your simulation to your data. Whenever you want to understand the system, rather than simply "treat" it, there is a need to go analyze the simulations to isolate the simple quantitative but approximate physics that is at the core of the phenomenon-- and that is what I find is all too often missing.
Well, I wouldn't say they are driven by the outcomes of complicated "black box" simulations with so much complexity that only a computer could keep track of it-- that's what no one would invest money in, given how fraught with peril such an approach would be! They are indeed theory driven, but it's more missing-theory driven. The missions are not just to add another significant figure to the model parameters, that would again be entering a realm of conceptually meaningless engineering adjustments to parameters that are quite likely to be missing physics at that level of precision. So I'd say the real goal of the missions is to look for any physics that we are missing-- physics that might have very simple and straightforward implications that do not require a mile of code to extract.
But is that fine-tuning done to be able to differentiate between conflicting outcomes of two groups doing black-box simulations, or to be able to distinguish between physical possibilities that have very straightforward, if not downright simple, implications?

__________________
Logic is the grammar of truth.

Meaning and absolute certainty are incompatible, and profound meaning and absolute certainty are profoundly incompatible.

The only thing intelligence is capable of is recognizing itself.