What Bureaucracies Remember

The W-2 form was designed for a worker with one employer, a regular salary, and a predictable annual income. It asks for an Employer Identification Number because it assumes there is one employer. It asks for wages, tips, and other compensation because these are the categories of income the form’s designers recognized. It does not have a field for income from selling homemade food out of your kitchen, or earnings from six different gig platforms across the year, or the cash paid by a neighbor for childcare on Tuesdays and Thursdays. These forms of income exist. They are simply not legible to the form.

This would be a minor administrative inconvenience if the W-2 were just a tax document. But the form’s categories travel. They get reproduced in rental applications, loan underwriting, benefit eligibility determinations, and visa applications. The question “what does this person earn?” is answered, in bureaucratic contexts, by “what does the form say?” — and the form only says what it was designed to say. People whose incomes don’t fit the template are not invisible, exactly. They are consistently misread, and the misreading has consequences that compound.

There is a sociology of forms that has not quite been written, though it exists in scattered pieces — in the work of scholars who have traced how census categories shape which populations are seen to exist, how medical intake forms encode assumptions about family structure, how application requirements filter out the people they were supposedly designed to help. What this research is circling around is something like bureaucratic memory: the way institutions preserve, through their paperwork, a picture of the world as it was understood at a particular moment by a particular set of designers. The form encodes that picture. Then it applies that picture, again and again, to a world that has kept changing without informing the form.

The people who built these systems were not, for the most part, trying to exclude anyone. They were solving a genuine administrative problem — how do you gather consistent, comparable information from millions of people? — and they made choices that seemed reasonable given what they could see. The categories they chose reflected the social world they knew. That world had more regularly employed, single-employer, annual-salary workers than the world that followed. For the people the form was designed for, it worked. The problem is that it kept being used for everyone else too.

What would it look like to design a bureaucratic form with genuine epistemic humility about what it cannot see? This is not a rhetorical question. It has practical answers: build in free-form fields alongside fixed categories; design for the actual distribution of cases rather than the modal ideal; treat the gap between what the form captures and what the person’s life actually contains as data worth collecting rather than noise to be filtered out. Institutions that have tried this tend to find that the gap is large, and that it is not randomly distributed. The form has a blind spot, and the blind spot has a shape — one that is worth studying if you are at all serious about the people you are trying to reach.