For SNAP systems, paying less was harder than paying nothing

The week

For several weeks in the autumn of 2025, the federal government was shut down, and one of the most immediate worries was whether families would be able to buy groceries in November. The shutdown that began on October 1, 2025 left November funding for SNAP in doubt. SNAP, the Supplemental Nutrition Assistance Program, is the federal food assistance program once known as food stamps. The U.S. Department of Agriculture (USDA) funds the benefits, and states run the program day to day: they decide who qualifies, calculate how much each household gets and load the money onto cards.

The first turn came in court. By October 31, federal courts in Rhode Island and Massachusetts had ordered the administration to use contingency funds, money set aside for emergencies, to pay benefits. On Monday, November 3, USDA told the Rhode Island court it would fund 50% of benefits. Late on November 5 it said that figure was an error, cut the reduction to 35% and sent states revised allotment tables. An allotment table is the official chart that sets the maximum monthly benefit for each household size; every state’s system uses it as the starting point for its calculations.

The ground kept moving. On November 6, the Rhode Island court ordered full November payments by the end of November 7, and some states began issuing them. Late on November 7, Justice Ketanji Brown Jackson granted an administrative stay of that order, a temporary pause that gives the Supreme Court time to consider a case without deciding it. USDA then moved to stop states drawing down full benefits, an effort a Massachusetts court blocked on November 10.

The shutdown ended on November 12 with a law that funded SNAP for the rest of the fiscal year. In the space of about ten days, state agencies had been told to prepare for no payment, half payment, a larger partial payment and full payment, with some of those instructions overlapping. Behind each instruction sat computer systems, card processors and call centers that had to turn a legal and political decision into an exact amount on every household’s card.

Take a single mother in a small city on the morning of November 7, 2025. She has two children and a SNAP card that is usually reloaded early in the month. She has heard on the radio that benefits will be cut in half, then that the cut will be smaller, then that a judge has ordered full payments. Her state’s website says one thing; a neighbor who checked her balance says another. She calls the benefits hotline and waits on hold. When she finally gets through, the caseworker is honest: they do not yet know what she will receive, or when.

The politics of that fortnight are well covered elsewhere. The narrower matter is what it revealed about the systems that deliver benefits. States were asked to pay a fraction of a benefit, then possibly the whole benefit, under instructions that changed within days. Stopping payment entirely would have been simpler, since every system already handles a month with no issuance. Paying a reduced, recalculated amount was the hard case. The strain showed in three places, each with a different owner: the benefit calculation logic, the issuance rails, and communication with recipients.

From federal rule to grocery checkout

A benefit passes through three main pieces of technology on its way from rule to grocery checkout, and each is run by a different party.

The eligibility system. Each state runs a large computer system that holds every household’s case: who lives there, what they earn, what they pay in rent and other deductible costs. Federal rules set how those facts turn into a monthly benefit. The eligibility system applies those rules and produces a figure for each household every month. These systems are often among the largest and most complex a state owns, and many are built and maintained with the help of a systems integrator, an outside technology firm hired to build, customize and run them.

The EBT processor. SNAP benefits reach households through electronic benefit transfer cards, which work much like debit cards. A small number of processors run EBT systems for the states. Each month the state sends the processor an issuance file, essentially a long list saying “load this amount onto this card on this date.” The processor holds the accounts and makes the money available.

The retailer and the recipient. When the household shops, the grocer’s card terminal checks the EBT account and deducts the purchase. From the recipient’s point of view, all that matters is the balance on the card and the date it is refilled.

The arrangement resembles payroll. The eligibility system works out each employee’s pay, the EBT processor is the bank that deposits it, and the agency’s website and call center are the HR helpdesk that answers questions about the payslip. In normal months the chain runs like clockwork. November 2025 was like being told, mid-pay-cycle, to pay everyone a percentage of their salary, then a different percentage, then their full salary, while employees phoned in asking what to expect.

Calculation logic: no setting for “partial”

State eligibility systems calculate SNAP benefits from household size, income and deductions, within federal rules. They are built to calculate the correct benefit, not a percentage of it. There is no box marked “pay 65 percent this month.”

USDA’s revision came as reduced allotment tables, which cut the maximum benefit. The normal formula then subtracts 30% of household net income, meaning income after allowed deductions. Many households therefore received less than 65% of their usual benefit, and some households with income could see the amount fall to zero. Interactions with minimum benefits, prorated first-month benefits (smaller amounts for people who joined partway through a month) and mid-month changes added edge cases.

The arithmetic shows why “a 35% cut” was misleading for so many families. To keep the numbers simple, take a purely hypothetical maximum benefit of $100 and a household whose 30%-of-net-income contribution comes to $40. In a normal month it receives $100 minus $40, or $60. If the maximum is cut by 35% to $65, the household receives $65 minus $40, or $25. Its benefit has fallen by more than half, against a headline cut of 35%. A household with no income, by contrast, would see exactly the 35% reduction. A household with a slightly higher income might see its benefit disappear. The same headline percentage lands very differently in each home.

A senior USDA official told the court that partial payments could take anywhere from a few weeks to several months, because nothing like them had been done before. Some states said they could move far faster. Each had to choose between speed and precision.

The underlying issue is configurability: how easily a system’s behavior can be changed by adjusting settings rather than rewriting its code. Many state eligibility systems are large, heavily customized platforms in which a change to benefit logic goes through a full release cycle, often with a systems integrator. A release cycle means designing the change, writing the code, testing it, getting approvals and deploying it. That process suits policy changes made through legislation and rulemaking, which usually come with months of notice. It fits poorly when the policy may change again before the release ships.

Agencies can reduce this risk in advance. A parameterized control can apply a percentage, a cap or an alternative formula across a caseload, with defined rules for edge cases. “Parameterized” means the system is built with a dial already in place: policy staff can set it to a value, rather than asking programmers to build a new dial in a hurry. Policy staff should be able to switch it on with recorded approvals rather than a code release. It needs testing against real caseload data every year, the way disaster recovery is tested. Contracts with integrators should treat emergency benefit changes as a defined service with a response time, rather than a change request priced after the fact.

A drill that takes days

Testing an emergency benefit control does not have to be elaborate. Once a year, policy, technology and finance staff take a copy of the live caseload into a test environment. They switch on the reduction with a made-up percentage and compare the results with an independent calculation for a sample of households, including the awkward ones: households near the minimum benefit, households in their first month, households with a change reported mid-month and households whose benefit would fall to zero. They then produce a mock issuance file and confirm that the EBT processor can accept it on a shortened timetable. Finally, they draft the notices and text messages that would go to households, and check that the portal can show each household its own figure.

The exercise takes days, not months, and it settles what mattered most in November 2025: how long it would actually take this agency, with this system and these partners, to deliver a partial payment correctly, as opposed to whether one is possible in principle.

Issuance rails: the EBT processor in the middle

Issuance files move from the state system to the processor on a schedule, often staggered by case number or last name across the month. Staggering spreads the load on stores and systems, so not every household shops on the first. Changing the amount, timing or sequence of those files at short notice depends on the processor’s capacity and contract terms as much as on the state. A state that wants to send a revised file tomorrow needs a processor that has agreed, in advance, to accept one.

November also exposed a problem few agencies had planned for. What happens if benefits are issued and the legal basis then changes? USDA’s weekend effort to stop full issuances ran into the reality that loaded funds can be spent within hours. The government itself told the Supreme Court it had no means to recoup money once transferred. Reversing benefits at the household level would be harder still and would damage trust in the program. Benefits issued under legal uncertainty are, in practice, final: once a family has bought groceries with them, no accounting entry can bring them back.

That has design consequences. Agencies should agree procedures with their processors for staged, delayed or supplemental issuance before they need them. A supplement, issued later to top up a partial payment, is far easier than a clawback, the term for taking money back. Issuance decisions during litigation should carry a documented legal and program sign-off, because they cannot be walked back. Reconciliation between the state system and the processor should be able to report, per household, what was issued under which instruction. When auditors, courts or legislators later ask what happened, that record is the answer.

Communication: telling people what they will get

The third strain fell on recipients. They needed to know whether they would be paid, how much and when. Agencies answered through websites, call centers, text messages and community partners while the answer itself kept moving. Because the reduced formula hit households differently, a single percentage in a press release was misleading for many of them.

Call volumes rose, and caseworkers fielded questions they could not answer. Food banks and community organizations became de facto information channels and absorbed demand. Agencies with established text messaging programs and clearly written status pages could update people quickly. Those relying on mailed notices and general press releases could not.

The remedies are simple, but they must exist before the crisis. Keep one authoritative status page per program, written for recipients and timestamped. Use direct channels, such as text and app notifications, that recipients have already opted into. Where the system can show each household its own expected amount through a portal or balance inquiry, do that rather than quoting averages. Brief call center staff and community partners from the same script at the same time, and say clearly what is not yet known.

Language access and retailer communication are easy to forget under pressure. Updates should go out in the languages the agency already uses for notices, at the same time as the English version. Grocers and farmers markets that accept EBT also field questions at the register, and a short bulletin through retailer associations reduces confusion at the point where recipients feel it most.

What rides on the calculation

For a governor, a human services secretary or a county administrator, these details decide whether families eat, whether the agency keeps public trust and whether the state ends up in court or in an audit.

People. SNAP households often have little cushion. Uncertainty about the amount and timing of a benefit forces hard choices about rent, utilities and food, and it pushes demand onto food banks that may already be stretched.

Staff. Caseworkers and call center agents bear the frustration of callers when they have no answers to give. Repeated weeks like that wear down a workforce that is already hard to recruit and keep.

Money and risk. Benefits issued in error, or under instructions later reversed, create reconciliation work and audit questions. Emergency changes negotiated after the fact with integrators and processors tend to be expensive. And a state’s payment accuracy is already a closely watched measure.

Legal exposure. In November 2025, agencies were acting under court orders, stays and federal guidance that changed within days. Decisions to issue or hold benefits had legal consequences, which is why they need documented sign-off from both counsel and program leadership.

Reputation. Residents rarely distinguish between federal, state and county roles. When a benefit is late or wrong, the agency whose name is on the card and the website takes the blame.

An agency built to pay less, later or in stages

Suppose a state human services agency has absorbed these lessons. Its eligibility system has a tested emergency control that can apply a percentage reduction, a cap or an alternative formula to the whole caseload, switched on by a named policy official with a recorded approval. Once a year the agency runs it against a copy of real caseload data and checks the edge cases. Its contract with its systems integrator defines emergency benefit changes as a service with a response time. Its EBT processor has agreed procedures for staged, delayed and supplemental issuance. Its counsel and program director have a standing sign-off procedure for any issuance made while litigation is pending. Its recipients have opted into text updates, and its portal shows each household its own expected amount. When the next disruption comes, it changes a setting, sends a revised file and texts its recipients the same day.

The agencies that struggle do the reverse. They treat benefit logic as fixed and every change as a software project, and assume the processor can accept any file at any time without prior agreement. They issue benefits under legal uncertainty without recording who decided and why, and publish a single percentage that does not match what most households will actually see. They brief the call center after the press release rather than before, and forget the grocers who face recipients at the register and the residents who read notices in languages other than English.

All three strains come from one assumption: that benefits flow on a stable monthly cycle in full amounts. Shutdowns, disasters and court orders break that assumption. Building for variability in calculation, issuance and communication is a resilience investment that agencies will use again.

How fast this agency could pay a partial benefit

1. If we were told tomorrow to pay a percentage of each household’s benefit, how long would it take, and who would have to change what?
2. Do we have a tested, parameterized way to apply a reduction or cap without a code release, and when did we last test it on real caseload data?
3. Has our EBT processor agreed procedures for staged, delayed and supplemental issuance, and could it accept a revised file at short notice?
4. Who signs off an issuance made while a court case is pending, and is that decision recorded?
5. Can each household see its own expected amount, or do we only publish averages?

With hindsight

The law that ended the shutdown on November 12 funded SNAP through September 30, 2026, so the most difficult scenarios did not play out at scale. Recipients were spared a prolonged period of partial benefits, and agencies were spared months of recalculation and reconciliation. The problem did not solve itself, though: the systems that strained in November are the same systems that will carry the next set of changes.

Those changes are already scheduled. The July 2025 reconciliation law expanded SNAP work requirements, raised states’ share of administrative costs from fiscal 2027, and ties a share of benefit costs to state payment error rates from fiscal 2028. Each of those changes runs through the same eligibility logic and issuance rails. Work requirements add new rules to the calculation and new information to collect from households. A larger administrative cost share puts pressure on the budgets that pay for system changes. Linking benefit costs to error rates means that every miscalculation, including those made in a rush, carries a direct price for the state.

Together, those shifts make configurability, accurate records and clear communication matters of state finance as much as of technology. Agencies with configurable benefit parameters, pre-agreed processor procedures and direct messaging channels are better placed for the next disruption. The test for any benefits platform is whether it can pay less, pay later or pay in stages under clear controls, and tell each household the same day. Agencies that build and rehearse those abilities now, while there is no deadline pressing, will find that the same work also helps them absorb the new work requirements, budget pressures and accuracy incentives that are already on the calendar.