Can you really separate the two like that?

You just did, dude. You just pointed out that the way our society works, for better or worse, is bridge designers that think you can do a 150' span with 4" deck (which can be done, just not the way they're thinking)

...and engineers whose life work is crushing their dreams.

It's unfortunate because dollars to donuts the dream-crushers actually give a shit about what the bridge looks like and, frankly, the designers do care that it works, they're just not given the body of knowledge necessary to make the relationship anything but adversarial. Screenwriting works the same way - every screenwriter is told to write what they imagine with no constraints, I guess so they can learn just how hard it is to engage a thousand people into spending a hundred million dollars on the 20,000 words you banged out alone in a Starbucks off Ventura blvd. Which isn't far off from your local governor:

As an extreme example, some local governor here came up with the term 'horsification' and a measure 'average amount of horses per acre' to quantify the demise of the countryside due to an increase in horses. You could totally do a time series analysis classifying horses with aerial photography and some ML, getting the data, but you can't deny that the data is meant to create a problem that wasn't there before.

So let's go with our horsification standard. I'm going to measure the horsification of a stretch of the Valle Grande in New Mexico at two points, historical and current because I'm lazy. For "current" I'm going to commission a satellite pass of the Valle and hire a grad student to count "horses." In order to distinguish between "horses" and "not horses" I'm going to need grazing rights records and animal registrations. Now that I've validated my model I'm going to give my grad student the last, best sat photos of the valle we didn't commission and have him count horses using his model. I now have two data points: horsification now and horsification, say, in 2013.

Now I need to quantify "demise of the countryside." Good luck with that but let's say I opt for methane emissions, tax rolls, traffic counts, I dunno. Stuff I can actually get for the year in question.

I now have two points of correlation that I think demonstrate that more horses = more "demise of the countryside." Thing is, all that data actually existed already - and if I want my argument to have any weight, I need to show my work. What I actually contributed to the conversation - a method for counting horses - may turn out to be a useful tool for other purposes. Meanwhile in the process of correlating horses with urban decay I've opened up an avenue of research, made some controversial statements and otherwise advanced the debate around ranching.

The article you linked literally argues that we shouldn't map shit because it hurts peoples' feelings. I think we can both agree that peoples' feelings are going to be hurt. My overarching point is that data doesn't hurt feelings. USE of data hurts feelings. "Horsification" is probably a bullshit metric, but it's a higher-order derivative of scalar data. Scalar data goes into lots of metrics, bullshit and otherwise, and when we all debate whether they're bullshit or not we all win.

You're going to have a tough time convincing me that simply measuring a value is EVER bad for society.