Every census is a portrait of what a state wants to know, which is not the same as what is true. The questions change more than the population does. A form from 1935 asks about literacy, livestock and the language spoken at home. A form from 1990 asks about refrigerators. A recent one asks about internet access and does not ask about language at all. Historians love censuses and distrust them in equal measure. The distrust is well founded. People lie on forms, but more importantly, forms make people lie by offering only certain boxes. A household with a complicated arrangement will be recorded as whatever the nearest available category is, and a hundred years later that category will look like a fact. There is a well known European example from the nineteenth century, where the number of people in one region reporting a particular mother tongue collapsed by more than half in ten years. No migration occurred. The wording of the question had changed. What survives is the answer, not the question, and this is the trap. Yet the alternative is worse. Without the census we would have almost nothing about ordinary people. The literate leave letters. The wealthy leave inventories. Everybody else appears in the record exactly twice, at birth and at death, unless a state counts them. The compromise most historians reach is to read a census as evidence of two things at once: something about the population, and something about the government doing the counting. The second is often the more reliable of the two. The 1935 form asked about livestock because livestock could be taxed. The recent one asks about internet access because access is now the thing that can be lacking. Both are honest, in their way, about their own priorities. Neither is honest about that on purpose.