You mentioned the issue here. More often than not, I care about the causal interpretation of the model. So I take it for granted. My mistake for not being clear.

Thanks both. Do you have any reference or example to share? Other comments seem to point towards "sensitivity analysis" kind of solutions.

Thank you both. I think the case I was referring to would fall into the "pre-specified" model category. But I like the article Tim posted above (I guess that would fall into the "etc"?).

I will certainly take a look at veridical data science.

However, sometimes I get the question "what if I have no idea of what the impact is going to be?"

I'll admit I don't have an answer to that. Not a polite one at least.