Four years early

42.2 million people on SNAP in May 2025. 36.6 million this May. Arizona, down more than half in twelve months. The Agriculture Department's newly released numbers show the biggest food-aid program in the country shedding more than 13% of its enrollment in a year as the work requirements in Trump's tax-and-safety-net law take hold.

In February the Congressional Budget Office projected those requirements would push enrollment below 34 million — by 2036.

A nonpartisan scorekeeper missing its own decade-long glide path by four years in the first year is not a rounding error. It is a sign the model priced the wrong mechanism. CBO modeled who would fail a work test. The rolls are being cleared by something faster and less discriminating than that.

The state distribution gives it away. Arizona's enrollment fell more than 50%. Georgia, Louisiana and Nevada fell more than 20%. No state labor market moves fifty points in twelve months, and no plausible story explains Arizona outrunning Nevada by more than two to one on employment grounds. What does vary that much between states is the capacity of the agency processing the forms — and AP reports Arizona's was overwhelmed trying to keep up with the changes. Tia Fields, who tracks safety-net policy at Invest in Louisiana, says the people she sees losing coverage mostly are not failing the work test. "A lot of it," she said, "is administrative paperwork."

That is the predictable result of the design. The new law didn't create a work requirement so much as extend an existing one to people who had been exempt: adults 55 to 64, parents of 14-to-17-year-olds, homeless people. Every one of those newly covered recipients now has to document compliance on a schedule, to a state office absorbing a rule change at the same time. The requirement is enforced through paperwork, and paperwork tests for paperwork.

The requirement is enforced through paperwork, and paperwork tests for paperwork.

The strongest case against reading it that way comes from the people who wanted the rule. Rachel Sheffield of the Heritage Foundation, which pushed for the tighter requirements, put the standard plainly: if people are leaving the rolls because they're working and moving forward, that's a step forward. She's right that this is the correct test, and some share of the 5.6 million did earn their way past the income line. Nobody should assume otherwise on the strength of a headline.

But that argument requires a number, and the number does not exist. AP's reporters say flatly it is too early to tell how many of the departures are people who found work versus people who missed a deadline or couldn't produce a document. USDA publishes enrollment. Enrollment falls the same whether you got a raise or lost a form. A policy sold on moving people into jobs is being evaluated on the one statistic that cannot distinguish a job from a filing error — and the states are about to get a second incentive pointing the same direction, since they'll soon owe part of the benefit cost when their payment-error rates run high.

Four years of projected decline arrived in one. That is either the fastest labor-market absorption of low-income adults in modern American history, or it is a documentation filter working exactly as a documentation filter works. The Heritage standard is the right one. Meet it: publish the employment number.