The example in Graeber's book was the government pays soldiers in some currency it invents, and then taxes the non soldiers in the economy the same amount. The soldiers need to eat (which is important as they might otherwise just give the coins to the civilians). So the transaction of food for money makes sense, and the civilians can pay their taxes and the soldiers can eat, and the arbitrary currency has a measurable value.
Ah - this makes sense, at least historically. With a tax that is more of a levy rather than something proportional to receipts or income, then it makes sense. But like the Mises Regression theorem, this seems more to relate to the historical origins of money as opposed to current usage.
