Start with the part that barely varies: what it costs a district to actually make the meal. USDA's most recent full cost study puts the average production cost of a school lunch at around $3.99, split almost evenly -- roughly 45% food, 45% labor, 10% everything else (supplies, equipment, utilities, administrative overhead). That ratio holds across most districts. It's a real input cost, not a policy choice, and it's the thing that stays roughly constant while what a family pays swings from $0 to $5.50 for the same tray. Three separate mechanisms, operating at three different levels of government, explain that swing -- and they stack on top of each other rather than replacing one another.

Mechanism one, state level: nine states pay for it so no family does, regardless of income. California, Colorado, and Maine started universal free school meals in the 2022-23 school year; Massachusetts, Michigan, New Mexico, and Vermont followed; Minnesota joined after that; New York is the ninth, phasing in for 2025-26. In those nine states, every public-school student eats free, funded by state money layered on top of the federal reimbursement. In the other forty-one, meals are means-tested against actual household income, not an abstraction most people carry around -- for a family of four in SY 2025-26, free at or below $42,107 a year, reduced-price (capped by law at 40 cents) at or below $59,922, full price above that. Those dollar figures are just 130% and 185% of the federal poverty line, respectively, but the percentage alone means nothing to a family trying to figure out which bucket they're in -- the dollar number is the one that actually answers the question.

Mechanism two, school level: a federal option that ignores state and district lines entirely. The Community Eligibility Provision (CEP) lets any individual school, not a whole state or even a whole district, serve free meals to every student on that campus once at least 25% of its students are already identified through SNAP or TANF -- no household applications required. The reimbursement is formula-driven: that identified-student percentage times 1.6, and a school at 62.5% or higher gets the full free rate for every meal it serves. This is why one school can be free and the school two miles away, in the same district and the same state, isn't -- it's poverty density measured building by building, not a district-wide or state-wide policy.

Mechanism three, district level: local price-setting, pushed upward by federal rule and real local budget pressure. Districts aren't free to set a low full price indefinitely -- the federal Paid Lunch Equity rule requires the average price charged to paying students to trend toward the free-meal reimbursement rate (around $4.60 a meal for SY 2025-26), which pushes prices up over time on its own. On top of that, actual local food, labor, and fuel costs, plus district-specific budget shortfalls, drive real district-to-district differences. New Jersey's 2025-26 maximum price schedule tops out at $5.00 (elementary) to $5.50 (high school). Broward County, Florida -- Fort Lauderdale -- is ending its universal-free-lunch pilot and raising full price for 2026-27 to $4.00 / $4.35 / $4.50 by grade band, a change the district ties directly to a $90 million budget gap, while reduced-price stays fixed at 40 cents because that number is set federally, not locally.

What's underneath mechanism three, in Broward's specific case: enrollment share, not just budget math. A district's federal reimbursement is paid per meal served, so a shrinking enrollment base shrinks that revenue at the same time fixed costs -- kitchen labor, equipment -- don't shrink with it. Broward's $90 million shortfall is tied to roughly 10,000 fewer enrolled students this year alone, and the district's own reporting names several separate causes: immigration enforcement (one board member attributes roughly 80% of the loss to it), declining birth rates, and increased use of private-school vouchers and school choice. That last cause is worth being precise about rather than assuming: opting a child out of public school doesn't require the public school being bad. Plenty of families in some of the country's best-rated public school districts choose private school for reasons that have nothing to do with the local public option's quality -- religious education, a specific pedagogy, family tradition -- so enrollment-share decline is real and it does drive the budget pressure behind mechanism three, but it isn't, on its own, evidence about public-school quality. The same mechanism shows up in miniature at the building level: USDA's own cost data found per-student meal costs actually rose at small schools (under 500 students) after they adopted universal-free programs, because kitchen labor and equipment don't shrink proportionally with a smaller student base -- a whole district losing enrollment feels the same squeeze at scale.

Read together: the three pricing mechanisms don't answer the same question. State policy (nine states) answers "does income matter here at all." CEP answers it building by building, based on measured poverty density, inside states where income still matters everywhere else. Local pricing answers a completely different question -- what does a family who doesn't qualify for either actually owe -- and that number is shaped by a federal equity floor pushing it up, a district's own budget reality pushing it up further, and, in districts with shrinking enrollment, a smaller base of meals to spread fixed costs across pushing it up again.

One number worth checking rather than assuming. A family in Fort Lauderdale heard on local TV that lunch runs $185 a month unless a student qualifies for the low-income discount. Broward's own published full price for 2026-27 is $4.50 a day at the high school level -- at roughly twenty school days a month, that's closer to $90, not $185. That gap is worth naming plainly rather than smoothing over: either the segment quoted a different fee entirely (a combined breakfast-and-lunch charge, a different district, an annual number misheard as monthly), or the number simply didn't survive being repeated. The real, checkable number is $4.50 a day.