SNAP's rolls fell 5.6 million in a year, and nobody can yet say how much of it is paperwork

The disagreement between an advocacy analyst and a think tank fellow is really a disagreement about whether the money comes back. Rachel Sheffield, a research fellow at the Heritage Foundation, which pushed for the stricter rules, framed the good outcome as people "leaving the welfare rolls because they're working and they're moving forward" ^[5]. That kind of exit is durable. Reporting by Fortune notes that some people losing coverage do qualify and are rejected for missing deadlines or not having documents to hand, and that it is too early to tell how many fall into each group ^[4].

The two categories do not behave alike downstream. An earnings exit removes a household from the caseload and leaves it out. Procedural denial produces churn: eligible, denied, reapplied, restored, with the timing set by agency capacity rather than by the labour market. The state spread is the tell. Against a national decline of 13.3 percent ^[1], Arizona lost 55 percent of its caseload in twelve months while its agency was struggling to absorb the changes ^[6], and Georgia, Louisiana, Nevada and Florida each shed more than 20 percent ^[7]. No plausible state-level income story runs four times the national rate in one year.

The retail arithmetic is worth doing even though the source data will not close it. The average monthly benefit is $344 per household, on cards usable only for groceries ^[8]. Applied to 5.6 million departed recipients, that is about $1.93bn a month, or $23.1bn a year ^[4]. Treat it as a ceiling, not an estimate: benefits are counted per household, and 5.6 million people occupy fewer than 5.6 million households, so the real withdrawal is smaller. It also lands with no geographic evenness at all, in a programme that feeds more than one in ten Americans, most of them below the poverty line ^[9].

Two mechanisms are still loaded. The expanded work requirement has taken effect in most of the country but does not start in some places until next year ^[14], and it now reaches people aged 55 to 64, parents of teenagers aged 14 to 17, and groups previously exempted, including homeless people ^[13]. Separately, from October 2027 states with payment error rates above 6 percent must fund part of the benefit cost, and advocates argue some will deny marginal cases outright rather than risk an error ^[12]. That provision pays states to be strict about documentation, which is the same lever that is already removing people.

Scale, finally, against the official path. The Congressional Budget Office in February had enrollment falling below 34 million by 2036 and did not expect this pace; May's count already matched what it forecast for 2030 ^[10]. That leaves roughly 2.6 million of projected decline for the remaining decade, against 5.6 million already gone in a single year ^[3]. Measured from the October 2024 peak of 43.3 million, the caseload is down 6.7 million, or 15.5 percent ^[2]. Anyone budgeting off the CBO glide path is working from a curve the programme has already left behind.