The message arrives with the serenity of a meditation app: “For your security, your session has expired.” You were forty minutes in. You had dug up the pay stubs, found your landlord’s legal name, answered the question about assets that required calling your sister. The form has forgotten all of it, and it does not apologize, because forms never do. Somewhere between the end of a shift and the arrival of a bus, a page refreshes and an hour of a person’s life is converted into nothing.
This is usually filed under “inconvenience,” the way a delayed flight is an inconvenience. It isn’t. When the form in question decides whether a household eats next month, a discarded hour of work is not a usability bug. It is a denial — one that no caseworker signed, no legislature approved and no agency will ever count.
Researchers who study this have a bloodless term for it: administrative burden. In a recent overview in the Journal of Economic Perspectives, Pamela Herd and Donald Moynihan divide it into three costs. Learning costs: figuring out that a program exists and whether you qualify. Compliance costs: the documents, the interviews, the portal. Psychological costs: the stress and low-grade humiliation of being processed. Their conclusion is blunt — these frictions “substantially limit eligible peoples’ access to public services they want, would benefit from, and are legally entitled to receive.” Legally entitled. The law says yes. The interface gets the last word.
If that sounds like an overstatement, consider an experiment. The SNAP intake process in Los Angeles required every applicant to complete an interview — not an interrogation, just a conversation. The county scheduled it; applicants waited for the phone to ring, often at work, often from a number they didn’t recognize, and missing the call could sink everything. A team of economists randomized some 65,000 applicants and let one group initiate the interview themselves, calling in when it suited them. Same eligibility rules, same program, same budget. Approvals rose by six percentage points. Early approvals doubled. Months later, that group was still more likely to be receiving benefits — by more than two percentage points. Nothing about who deserved food had changed; only the choreography had. The study, published in the American Economic Journal: Economic Policy, names the interview itself as a key barrier to take-up — a polite way of saying that some of the gap between “eligible” and “enrolled” is a phone call arriving at the wrong time.
The cheapest cut in government
Why does this persist? Because burden is the most cost-effective policy instrument ever devised. Cutting benefits requires a vote, a headline, a backlash. Declining to fix the website requires nothing at all, and every friction arrives dressed as a virtue: security, integrity, fraud prevention. None of it appears in a budget. And the filter sorts with eerie precision — not by need, but by stamina, literacy, free time and reliable Wi-Fi.
A nationally representative survey of roughly 2,500 low-wage American workers, conducted through Washington University’s Center for Social Development, measured the sorting directly. Among respondents who had dealt with SNAP, 51 percent reported at least one administrative burden; for Medicaid, 43 percent. Learning burdens were the most common — people could not find out what they were owed. The damning part is the correlation: the SNAP applicants hitting the most friction were the most likely to be food-insufficient, and the Medicaid applicants hitting it were the most likely to report chronic illness and care they couldn’t afford. The hurdle stands highest exactly where the stakes do. And where agency staff carried heavier workloads, the study found longer waits, more processing delays, more procedural errors. Underfunding converts, mechanically, into rationing.
You can measure the burden with free software, because of course you can. In a 2025 brief, the researchers Zhang, Roll and Despard ran state SNAP websites through Google Lighthouse — the same automated audit a startup intern runs on a landing page — and found that the sites varied widely in speed, accessibility and mobile design. The differences were not cosmetic. Faster-loading application pages went with lower reported burdens of all three kinds, and better design tracked with better timeliness and fewer payment errors in USDA quality-control data. Silicon Valley A/B-tests the color of a checkout button; much of the safety net has never been through an audit that costs nothing and finishes in ninety seconds.
The computer says no
The obvious fix — more automation — turns out to have its own paperwork. Britain’s Universal Credit is digital by default: payments calculated automatically, earnings pulled from real-time payroll data, a claimant’s life administered through an online journal. Automation was meant to abolish the runaround. When the researchers Bennett, Currie and Podoletz interviewed 44 recipients for the Journal of Social Policy, they found it had mostly relocated the runaround. Payment errors now arrived at machine speed. The system shared too little about what it had calculated, or why. And with face-to-face contact engineered out, there was no one left to argue with — only a record of decisions made by a process no one could cross-examine. The discretion didn’t disappear. It moved upstream, into design choices made by people who will never watch a session expire.
That is the part worth sitting with. We conduct our arguments about the safety net in public — who qualifies, where the line sits, what the country owes — as if eligibility were decided in legislatures. But some of its operative rules are written in session lengths, interview-scheduling logic and page-load times, and those rules answer to no one. A procedural denial looks like nothing: a missed call, a reset form, a case closed for “failure to complete.” It produces no vote and no signature. It just produces less.
The standard is sitting right there. Programs get audited for fraud down to the dollar; the forms could be audited for the eligible people they discard, and the tools to do it are free. Until the second count matters as much as the first, the session timeout is not a glitch. It is the government, deciding — off the books, at scale, one expired hour at a time.