The application came home in a backpack, or it came as a link in an email, and it wants to know how many people live in your house and what they bring in. Last year nobody asked. The year before that nobody asked either. Your kid ate breakfast and lunch at school, it cost nothing, there was no form, and now there is a form.

No school board voted to stop feeding the children. Nobody ran for office on it. What moved was a percentage, and the percentage is written into federal regulation, and somebody counts it on the first day of April.

The machine has three moving parts, and they run in this order.

Your child was signed up without you filling out anything

The first part is the one almost no parent knows about, because it was built to be invisible.

7 C.F.R. § 245.6 carries the title "Application, eligibility and certification of children for free and reduced price meals and free milk." Paragraph (b) is called direct certification, and it opens like this: "In lieu of requiring a household to complete the free and reduced price meal or free milk application ... the local educational agency must certify children as eligible for free meals."

The regulation is not freelancing here. Congress wrote the rule first. 42 U.S.C. § 1758(b)(4) says a child in a SNAP household is certified for free meals "without further application," and § 245.6(b)(1) tells the district how: by data match. Not by asking. By running one list against another.

Must. Not may. If your household gets SNAP, the school district is required to match its enrollment list against the state's SNAP records and certify your child for free meals with no application from you at all. The same rule tells districts that direct certification "must allow for notifying parents that their children have been determined eligible for free meals or free milk ... and that no further application is required." Households on TANF or the Food Distribution Program on Indian Reservations can be certified the same way, and so can children who are homeless, migrant, runaway, in foster care, or enrolled in Head Start.

The match is not a one-time thing either. Paragraph (b)(3) makes districts run it at the start of the school year, again three months later, and again six months after the first. Three passes a year, minimum, every year, forever.

Here is the part that matters later, and it is the whole reason this piece exists. A data match runs in both directions, and nobody has to do anything for it to run the wrong way. No clerk types a child out. When a household comes off SNAP, it is simply not in the file the next time the district pulls the match, and the certification that file was carrying ends right there. No letter goes to the parent. No letter goes to the school. A child who was certified in October because a computer said so is not certified in February because the computer stopped saying so, and the only visible artifact of any of it is a form in a backpack.

The count of children the government already knows about is what everything downstream is built on.

Twenty-five percent, counted on the first of April

The second rule decides whether a school feeds everybody.

7 C.F.R. § 245.9(f) is the Community Eligibility Provision. A school, a group of schools, or an entire district can elect it, and then for up to four straight years it serves free breakfast and free lunch to every enrolled child and collects no applications from anybody. Not "no applications from poor families." None. The regulation is blunt about it: a participating district "must not collect applications for free and reduced price school meals on behalf of children in schools participating in the community eligibility provision."

To elect it, a school has to clear one number. The regulation calls it the identified student percentage: the count of kids already certified without an application, divided by the count of kids enrolled, times 100. Those are the SNAP and TANF and FDPIR kids from the paragraph above, plus children certified through Medicaid in the states where USDA has approved that match, plus the foster, homeless, migrant, runaway and Head Start kids. SNAP is the only one of those a district has no choice about. Section 245.6(b)(1)(i) says local educational agencies "must directly certify" children in SNAP households; every other category on that list is a "may" under § 245.6(b)(2). It is still not the only input, and this piece is not going to tell you the identified student percentage is SNAP, because it is not. It is not a poverty rate either, and it does not pretend to be. It is a headcount of children the state's own computers have already flagged.

The floor is 25 percent, and the regulation fixes the date it is taken: "as of April 1 of the school year prior to participating." The paperwork electing CEP for next year is due to the state by June 30.

So the school year that just started was decided in April, on a count nobody outside a district data office watched, and locked in by a form filed in June.

Now go looking for that 25 percent in the law, and you will not find it.

Congress wrote 40. The statute, at 42 U.S.C. § 1759a(a)(1)(F)(viii), sets the identified student percentage at 40 and then hands the Secretary of Agriculture the pen to write a lower one. The 25 that is actually in force lives one level down, in 7 C.F.R. § 245.9(f)(3)(i), and it got there through a USDA final rule published September 26, 2023, 88 Fed. Reg. 65778, RIN 0584-AE93, effective that October 26. Its stated reason was to give schools "greater flexibility to offer meals to all enrolled students at no cost when financially viable."

That is the title of this piece, sitting in a citation. Nobody voted on 25 percent. An agency wrote it, using authority Congress handed over, and a future Secretary can walk it back to 40 by exactly the same route: a proposed rule, a comment docket nobody in a cafeteria reads, a final rule, a date. No vote is taken on the way down and no vote is taken on the way back up. The one thing you get either time is a comment period, which is more than you get for the other half of this story.

And when financially viable is doing real work in that sentence. That phrase is where the second number hides.

Sixty-two and a half is where the money runs out

Because 25 percent is the line for allowed. There is a second number nobody prints on a flyer, and it is the line for free.

Federal reimbursement in a CEP school is not one flat check. Paragraph (f)(4)(v)(A) of the same regulation sets it by formula: take the identified student percentage, multiply by 1.6, and that is the share of meals Washington pays at the free rate. Everything left over, up to 100 percent, gets the paid rate, which is small change. USDA printed the math in the 2023 rule in one line: % Meals reimbursed at Federal free rate = ISP × 1.6.

Run it. An identified student percentage of 62.5 times 1.6 is 100. At 62.5 percent, every tray is federally funded and the district writes no check. At 45 percent, about 72 percent of trays are covered at the free rate and the district eats the rest. At the 25 percent floor, 40 percent of the trays are covered and sixty cents of every dollar of food is a local problem.

And the regulation is explicit about who pays that gap. § 245.9(f)(4)(vii) requires the district to cover it "with non-Federal sources of funds."

So there are two thresholds and they are not the same threshold. Twenty-five percent is where a school stops being eligible. Sixty-two and a half is where a school stops being free. The distance between them is a district budget, and a district budget is a school board deciding whether to cover lunch out of the same pot that buys buses.

Call the two of them what they are. The cliff is a school falling under 25 percent and losing community eligibility outright, and it is the one that makes the news. The squeeze is quieter and it arrives years earlier: the share of trays Washington pays for at the free rate slides down a point at a time while the bill for the remainder lands in the general fund, and nothing about it is announced. Districts leave community eligibility over the squeeze long before the cliff ever gets near them. Which is why a school can stop feeding every child free while sitting nowhere close to the floor, and why the next section is about a district that was never at risk of the cliff at all.

Miami-Dade went from 64 percent to 45.1, and the district named a different reason

Florida has the clearest live example, and the district's own explanation of it is not the one you would expect.

Miami-Dade County Public Schools, the third-largest district in the country, ran districtwide community eligibility for a full four-year cycle and did not renew it for 2026-27. WLRN reported on July 23, 2026 that the district's directly certified share had fallen from 64 percent four years earlier to 45.1 percent. Local 10 reported the next day that families at the affected schools could start applying August 10, and that students would keep getting meals at no charge for the first 30 days while the district worked out who qualified. K-12 Dive reported on July 31 that 121 of the district's 500-plus schools kept universal free meals on their own numbers.

Miami-Dade did not blame SNAP. The district's chief operating officer, Luis Diaz, wrote it in a July 30 post, quoted by K-12 Dive: "changing demographics, fewer newcomer students enrolling in our schools, and other enrollment trends have altered the federal eligibility calculations." That is a real explanation and it deserves to be taken at face value. Who shows up to enroll changes an identified student percentage all by itself, and nobody has to lose a benefit for that to happen.

And 45.1 percent is nowhere near the 25 percent floor. Miami-Dade was never barred from community eligibility. It was looking at the other number, the one where federal money stops covering the whole tab, and it decided it could carry 121 campuses and not the rest. WLRN described the 62.5 percent line as the federal threshold the district fell below. In the regulation, 62.5 is not an eligibility rule at all. It is just where 1.6 times the percentage hits 100.

How many Miami-Dade schools lost universal meals is a number the coverage does not agree on, so it is not in this piece.

315,666 Florida households came off the match in twelve months

One district is an anecdote. The state number is the thing to hold on to, and it comes from the agency that runs the program.

USDA counts people and households on SNAP state by state and month by month, and publishes the whole series as one archive of fiscal-year workbooks. Florida had 2,992,240 people on the program in February 2025. In February 2026 it had 2,412,302. That is a drop of 579,938, or 19.4 percent, against a national decline of 10.0 percent over the same twelve months. Florida is about a seventh of the entire national fall by itself.

Households are the number that actually drives a school district's spreadsheet, because direct certification matches households and not headcounts. Florida went from 1,686,934 SNAP households to 1,371,268. That is 315,666 households out of the direct-certification match, each one of them carrying whatever school-age children it carried, every one of those certifications ending the way the last section described: quietly, in a file, with no letter to anybody.

Now the discipline, because this is exactly where the coverage gets loose and this show does not.

Those numbers are facts. What caused them is not in the table. Redeterminations that were paused and then restarted, state administrative practice, ordinary county-level churn, and the new federal work rules are all live candidates, and no primary document in hand apportions the decline among them. Nobody at USDA has published a finding that says which. So this piece will not hand you one. The fall is the receipt. The reason is an argument, and anybody selling you a clean version of it is selling.

It was not a rule. It was a law, and it has section numbers.

Most of what you have read about the SNAP changes calls them a rule, or a policy, or a cut. The word matters more than it looks, because a rule is a thing you get to comment on and a statute is not.

Public Law 119-21 was signed July 4, 2025, and it rewrote SNAP by direct statutory amendment at sections 10101 through 10108, with the eligibility changes at 10102 and 10108. Section 10102 raised the top of the work-requirement exemption from 55 to 65, and cut the dependent-child exemption from a child under 18 to a child under 14. Section 10108 narrowed eligibility for lawfully present immigrants. Section 10105 is the one that comes back later in this piece: it adds a state cost share beginning in fiscal 2028, tied to how accurately a state pays benefits.

There is no rulemaking implementing any of the eligibility cuts. A search of the agency's published documents since July 2025 turns up none: no proposed rule, no comment period, no Federal Register citation. That is not an oversight. When Congress writes the eligibility text into the statute itself, the agency does not have to ask anybody anything, and there is no docket for a parent, a school district or a state to file into.

Hold that against the paragraph about the 25 percent floor. The floor that decides whether a school can feed every child is an agency number, movable by a Secretary, with a comment period attached. The rules that decide who stays on SNAP in the first place are a statute, moved by a vote most people did not know was about their kid's cafeteria. One of those two things had a public participation step and it is not the one you would guess.

The Congressional Research Service laid the eligibility changes side by side in report R48755, published December 11, 2025. Prior law applied SNAP's three-months-in-thirty-six time limit to able-bodied adults without dependents aged 18 to 54. The new law extends it to adults 55 through 64, and to parents whose youngest child is 14 or older. It also struck three exemptions that were on the books until 2030: veterans, without regard to disability status. Homeless individuals. Former foster youth who had been in care on their eighteenth birthday. CRS records the SNAP provisions as effective on enactment, with USDA guidance requiring states to have them running by November 2, 2025.

The connection between that law and that cafeteria is real. It is also worth being precise about what kind of claim it is.

No USDA document says that cutting SNAP will end universal free meals at your child's school. Nothing published by the agency draws that line, and this piece is not going to pretend one does. The two rules above give you math you can run on a napkin. SNAP is the one direct-certification match federal regulation makes mandatory, and every other category is optional. Direct certification is the numerator of the identified student percentage. The identified student percentage is the number a district takes to the state on June 30. Shrink the first and the last one falls, on its own, with no vote and no announcement, and it falls hardest at the schools already sitting near a line. That is a reading of two federal rules against each other. It is not an agency finding, and calling it one would be a lie told in a direction I already lean.

What is still missing is the school number, and it is worth naming the hole precisely now that the SNAP side of it is filled. Nobody has published a count of how many Florida schools lost community eligibility for this school year, or how many students were in them. That count would settle the argument in either direction and it is not traced, so this piece does not estimate it. The national counts of community eligibility schools that circulate in coverage are the Food Research & Action Center's, an advocacy organization's, not USDA's, and USDA publishes no national total at all. They are not in this piece either, and when you see one, check whose it is before you repeat it.

The rest is not complicated. A seven-year-old does not have a work requirement. A seven-year-old cannot be a veteran who lost an exemption, or a fifty-eight-year-old who aged into one. The adults who lost something under the new law and the children who lose the tray are not the same people, and the children who never lost anything at all get the form too, because the form comes back for the whole building at once.

The strongest version of the other side is a number, and it is a real one

Nobody defending this law says the quiet thing. What they say is accuracy, and they have math to say it with.

SNAP's national payment error rate was 10.62 percent in fiscal 2025, in USDA's own published rates. It has been above 10 percent in every year the agency has published a national figure since fiscal 2021: 11.54 percent in fiscal 2022, 11.68 in 2023, 10.93 in 2024, and 10.62 in 2025. There is no published national rate at all for fiscal 2020 or fiscal 2021. Washington funds the benefits. The states run the eligibility work, hire the caseworkers, build the systems and make the mistakes. What is not true is that states had no exposure to their own errors. 7 U.S.C. § 2025(c) has required the Secretary, for fiscal 2004 and every year since, to set a liability amount for a state whose error rate runs above 105 percent of the national measure for a second straight year, worked out as the state's whole year of benefits times the points it sits above 6 percent times 10 percent. USDA's own quality control page says what happens to that amount: a state can pay it to USDA outright, or settle it by putting half into fixing what caused the errors, with the other half held "at-risk" and collected only if a third bad year follows. What that liability never reached is the benefits themselves. Section 10105 changes that: starting in fiscal 2028, a state's error rate sets the share of its own benefit costs it has to carry. Make the party that errs pay for the error. As a design principle that is not stupid, and a show that pretended otherwise would be doing you a disservice.

Here is the answer, and every piece of it comes out of the same document.

That rate does not measure fraud. It measures dollar accuracy: how far the benefit issued sat from the benefit the household was actually owed, in either direction. And 1.33 of those 10.62 points are underpayments, money states failed to give people who qualified for it. Nothing in § 10105 tells an underpayment from an overpayment when it computes what a state owes. A state that shorts a family is charged on the same schedule as a state that overpays one, which means the cheapest way to move the number is not to be more accurate. It is to be slower and stingier at the front door, where the errors that get counted are the ones nobody is looking for.

The tiers are cliffs too, not slopes. A state at 9.99 percent owes 10 percent of its benefit costs. A state at 10.01 percent owes 15 percent, on the whole allotment, for two hundredths of a point of difference.

Florida's fiscal 2025 rate is 12.97 percent. That is the top tier, and on the 2025 rate it means a 15 percent share of Florida's own benefit costs beginning in fiscal 2028. Section 10105 also writes in a delayed start, and it names the line. If a state's fiscal 2025 error rate "multiplied by 1.5 is equal to or above 20 percent," the section pushes that state's start to fiscal 2029. Run Florida's: 12.97 times 1.5 is 19.46, which misses the 20 by half a point. On its 2025 rate, Florida does not get the extra time.

That is the settled part, and it is the only settled part. The same clause carries a second test on the fiscal 2026 rate, which pushes a state to fiscal 2030 instead, and USDA has not published a fiscal 2026 rate. Any fiscal 2026 rate of 13.34 percent or higher clears that bar, and nobody outside the agency knows yet whether Florida's does. The law also lets a state "elect to use the payment error rate of the State from fiscal year 2025 or 2026" when the fiscal 2028 share is calculated, so which rate governs is Florida's choice and not a settled fact either. What a state does about a bill like that is the next thing worth watching, and this piece is not going to pretend it already knows.

30 operating days, and one date in April

Four moves, each one with a date attached to it.

Fill out the form, even if you are sure you will not qualify. Section 245.6(c)(2) lets a school carry last year's status for the first 30 operating days of the year and no longer. After that, a child with no application on file gets charged. And the identified student percentage is not the same test as free-and-reduced eligibility, so plenty of families who were never on SNAP still land under the income line.

If your household starts SNAP mid-year, you do not have to reapply for meals. Paragraph (b)(3) requires the district to run the match again three months into the year and again at six. That is the rule to quote if a school office tells you otherwise.

April 1 is the count and June 30 is the deadline. If you want a school board to look at whether it can hold community eligibility by grouping schools together, the meeting to say so at is in the spring. Grouping is allowed on purpose: § 245.9(f)(3)(i) lets an individual school sit below 25 percent inside a group whose average clears it. A district that files nothing in June has decided by default.

Ask the state how it is doing on its own benchmark. Section 245.13 requires every state agency to directly certify 95 percent of the children in SNAP households, and a state that misses has to file a continuous improvement plan with USDA. When a state's matching misses children who qualify, a school's percentage drops without one family losing a benefit. That is a state-level failure with a state-level owner and a paper trail attached.

Florida already prints a list of what a family may not buy with cash assistance and charges those same households extra for being poor. People learn fast that none of it is theirs to touch, and they keep not asking long after the rule that taught them is gone.

Two federal regulations, one multiplication, one date in April. A spreadsheet crossed a line in a district office, and a cafeteria that fed every child who walked into it started sorting them by what their parents wrote on a form. Nobody had to vote for it, and nobody has to answer for it either.

Come back for the rest

We do this every week. The official reason on top, the math underneath, every number traced back to the regulation it came from so you can go read it yourself. Pour something cold, pull up a chair, and let us send you the receipts before the next round of nonsense clears your feed. Subscribe to The Long Pour. The free newsletter where the week's little robberies and the one big one land in your inbox together.

Y'all come back, now.


The Receipts

Every claim above traces to a document you can open yourself, with the access date on each line.