It's weird that this article doesn't even attempt to grapple with the reason why governments can't run big policies as experiments and cancel them if they fail:

Every time the government hires a group of people to do X, that automatically creates a class of people who:

1. Depend on the government continuing to do X for their livelihoods

2. Are experts in X and know far more about it than any government official

This creates an automatic constituency that will fight tooth and nail to keep X going no matter what. And step one of that fighting will be to make sure that the official report answering the original research question "Did X work?" will never be a clear "No."

And God help any politician that ignores the official report and cancels X anyway. Now the problem that X was intended to solve is entirely their fault, and there's an army of X experts running to every media outlet in the country making sure the general public knows it!

Your argument hinges on deciding what "works" means.

You're saying it "works" for some people but maybe not in a way or scale that was originally intended for it to "work".

But isn't that how all learning in a complex system takes place? Doesn't mean you need to rush to cancel something?

What's wrong in saying it doesn't do what we thought it would but it does do something useful? Maybe adjust further investment based on results obtained. Ignore sunk cost.

Please give examples of these agencies. What you list seems more like how business captures control via 'knowing more than government officials' then how technocrats do. At least that's what conservative politicians tell us as they defer all oversight/domain knowledge to private industry and private industry 'best' practices. Are you also saying that government should stop that practice and that is is also a trap of some kind? Or is it just a trap when it's for the betterment of society?

this is based more on assumption than anything else i think. this is something i hear folks yell very loudly about all the time but actual examples seem to be harder to come by.