Senator John Kennedy of Louisiana is laundering his cost objection with an Easter-bunny insult.

In June, at a Senate hearing on affordability, Kennedy said a federal job guarantee would cost trillions of dollars and was impossible. When Dr. Julie Margetta Morgan, president of the Century Foundation, told him she supported a federal job guarantee, he asked, “Do you believe in the Easter bunny?” The first claim needs a score. The second needs a design. The third is not fiscal analysis. It is a cost objection without a cost estimate, and it is an insult to the people the program would employ.

Here are the numbers. Representative Ayanna Pressley’s September 2026 resolution, reintroducing the federal job guarantee she first proposed in 2021, would establish a federally funded, locally administered program paying participants at least $25 an hour. The wage would be adjusted regularly to reflect the rising cost of living. Work would include emergency preparedness, disaster relief and recovery, childcare, elder care, and public art.

The current-law baseline—the budget path if current law remained in force—contains no federal job guarantee and no guarantee spending. At 40 hours a week for 52 weeks, the $25 floor pays $52,000 a year before administration, training, equipment, materials, payroll taxes, and benefits. The program cost is the wage bill: participants, hours or duration, and wage rate. Add the operating costs. Subtract avoided unemployment, disability, and early-retirement benefits. Add taxes and social-insurance contributions from newly employed workers. CBO would score the result over its ten-year budget window, with gross costs and offsets shown separately.

The accounting is not difficult because the social question is difficult. Wages and payroll taxes are costs. Administration, training, equipment, and materials are costs. A benefit no longer paid is an offset. Taxes and social-insurance contributions from newly employed workers are receipts. Better health, dignity, and useful public output are real benefits, but they improve the budget score only when they change federal spending or receipts. The rest belong in the program-effectiveness analysis.

Max Kasy’s accounting is useful because it lists offsets rather than pretending the wage is the entire ledger: unemployment benefits, disability or early-retirement benefits, taxes, social-insurance contributions, and useful work. Each offset depends on eligibility rules, replaced benefits, displacement, and enrollment. Do not book a saving because an economist says “maybe.” Do not book payroll and income taxes as new revenue without modeling the private work displaced. The score is how a claim survives contact with arithmetic.

Anyone who has read the Oxford Institute for Economics and Democracy’s release knows what the Austrian job-guarantee experiment reports: participants experienced social and psychological benefits, including higher well-being, and the program eliminated long-term unemployment among participants. That is evidence, and it matters. It is not a score for a nationwide program.

The next questions are administrative. Who hires? Who supervises? What counts as additional work rather than a transfer from another public program? How are wages set without displacing private workers? What happens when a local government lacks the capacity to administer the program? These are not details after the argument. They are the argument.

Pavlina Tcherneva of Bard College makes the welfare point: a good job is transformative, and the health costs of precarious and long-term unemployment turn employment from a cost into an investment. Properly designed, a guarantee is an automatic stabilizer: a government program that expands when private employment contracts and contracts when private hiring recovers.

A national guarantee must also supply national rights: the right to an offer, the wage floor, nondiscrimination, and benefit rules that do not punish people for taking work. Local entities and private employers can administer the work. They cannot be left to invent the rights of the people they employ. Federalism is not an exit.

The design must include an explicit phase-down when private hiring recovers, measured outputs for each local program, and a clear rule for benefits when workers move into private employment. Without those controls, the promise is not fiscal discipline. It is an open-ended claim on the Treasury. The purpose is not merely to occupy people. It is to produce public services while income continues.

The Cleveland pilot supplies a mechanism with a limit. Universal Basic Employment is launching a five-worker pilot that subsidizes private-sector wages rather than creating public jobs from scratch. Five van drivers at an existing transportation company currently make about $15 an hour. The pilot would lift their pay to just over $27, roughly the level at which most safety-net benefits phase out, with the employer and the program sharing the cost. The pilot is projected to cost almost $500,000 for one year.

Five workers cannot establish a national program. They can establish the accounting: the employer claims a labor-cost saving, the worker claims more income, and the program pays the difference. The pilot must also show that the hours are additional and that the subsidy is not buying a job that would have existed anyway. It shows how government can use private employers as delivery channels without surrendering the federal right, the wage floor, nondiscrimination, or benefit rules.

The rival instrument on the table is a direct tax on large AI companies. Representatives Greg Casar of Texas, Valerie Foushee of North Carolina, and Sara Jacobs of California would tax those companies on the higher of token value or AI-product revenue, with the rate rising as unemployment rises. The trigger is politically attractive: the bill scales itself to the crisis.

But scale the bill to the crisis and the base moves with it. If investors conclude that artificial intelligence is a bubble and pull their money out, the revenue base contracts at exactly the moment the guarantee’s claims expand. A financing mechanism that shrinks in the recession is not a financing mechanism.

Product revenue is the plainer tax base. Token value requires an accounting rule Congress has not supplied. The score must show who pays, how much is collected, and whether the burden passes through to prices, wages, or investment. The AI industry should pay for insurance against the displacement its technology may create. Its levy cannot carry the guarantee alone.

A Pew survey found majorities in 34 of 37 countries expect AI to reduce jobs. That finding justifies preparing an instrument before a crisis. It does not determine the instrument’s cost.

Economists have urged immediate action on AI’s economic risks. The case for preparation is no longer confined to predictions from technology executives. A guarantee enacted after layoffs begin is not prevention. It is emergency response at peak cost.

Put the two designs on one scale: gross cost per sustained additional job, whether the job would exist without the subsidy, administrative burden, displacement of other benefits, crowding out of other workers, service output, and the rate at which participants leave for unsubsidized employment. The Cleveland pilot can be scored on every one of those measures inside a year. A federal guarantee must build the same measurement apparatus before it launches. The phase-down rule, the output measures, and the exit path into private employment are not extras. They are the program.

Supporters must obey the same accounting. The social value of useful work does not make a federal cost vanish; it belongs in the analysis of program results. Opponents must obey the same accounting. A benefit no longer paid is an offset; a benefit that improves well-being but leaves the federal ledger unchanged is not. Gross spending rises when unemployment rises and participants enter. Offsets also change. The score must show both.

Congress should establish a federal job guarantee, federally financed and locally administered, with a $25 cost-of-living-adjusted wage floor, automatic expansion as unemployment rises, automatic contraction as private hiring recovers, and explicit national rights. CBO should score its net federal cost under current law over the ten-year window; GAO should test whether the promised services appear and whether health outcomes improve. Stable, broadly collected revenues should lead the financing. Any AI levy should supply supplemental revenue, not carry the guarantee alone.

Kennedy’s contribution to the June hearing was an Easter bunny. The required work is a table, a current-law baseline, and a ten-year projection.

The guarantee is the point; the score makes it real. The score is the test, not the verdict. The job guarantee is the policy. The score is the score. Build the guarantee on it.