I disagree. I understand the issues you're raising, but there is a key difference as far as we, members of the public looking at science, are concerned. The process of building models can't really be critiqued by non-scientists; but the process of verifying the predictions of models can, because predictions, at least predictions that non-scientists care about, can always be boiled down to something that anyone can test. For example, if you want to test whether relativity works, just use the GPS on your phone: if the location it tells you you're at is correct, you've just verified predictions made by relativity.
It is true that scientists do not always boil down their predictions to such simple tests, but as members of the public, our response to that is easy as well: if a scientist can't boil down the predictions of their model to a simple test like that, don't believe the model. In other words, we as members of the public can give incentives with regard to predictions that we can't give with regard to model building, or more generally scientific research as a whole.
> you can always either make the model vague enough or shape, manipulate, skew, explain away the subsequent measurements enough to get a "good predictive track record".
You can only fake tests that aren't real tests anyway. You can't fake something like GPS working: either it consistently tells you where you are correctly, or it doesn't. So that's the kind of tests that should be required before members of the public accept the claims of any scientific model.
That's neither here nor there though.
Those models might be abstract (in the sense that they're not concrete object) but they are still an example of concrete practice of science -- they were thought and formulated into specific words, measurements and papers.
In other words, models come to life by the concrete practice of science by humans (with all its faults). They don't manifest by some abstract science.
>Those models are science separate from how science is practiced as a human social endeavor.
It's the opposite.
Those models were built not "separate from how science is practiced as a human social endeavor" but exactly by people practicing science as a human social endeavor. And they were shaped by exactly the things you say they were not: who built them, what credentials they had, their beliefs and imperfections (e.g. lack of statistics knowledge), etc.
For example, if the scientists (actual persons) creating a model is greedy, they can create a model that advances their financials (rather than science) by optimizing for grants or publicity or "fancy results" or sponsors, against accuracy.
One might still agree, sure, but the model can still be verified independently of those bias and personal faults etc.
But that would be wrong too. Because there's no abstract verification. Verification also happens by actual people, as an actual process, and prone to the same issues: slopiness, greed, shallow peers merely linking to a paper they don't understand well, ideological biases, simply not caring, and so on.
They're not the abstract idea of science.
>If the experts are basing their expertise on models with a good predictive track record
Even that is not some fool-proof criterium.
Because you can always either make the model vague enough or shape, manipulate, skew, explain away the subsequent measurements enough to get a "good predictive track record". It's much easier in science than it is in sports bets where a team either wins or doesn't.
See, the verification of those predictions also happens by humans (with all the above issues).