We ran the banks' algorithm for a real person. The math said: the system is broken, not her.

The Setup

Banks use quantum computers to optimize portfolios. HSBC optimizes bond trading. Lloyds detects fraud networks. Goldman Sachs prices derivatives. They all use variations of the same algorithm — QAOA (Quantum Approximate Optimization Algorithm) — on IBM's quantum hardware.

We asked a different question: what happens when you use the same algorithm, on the same hardware, for a person instead of a portfolio?

Meet Maria. She's not real — she's a composite based on Brookings Institution data showing 6.1 million U.S. workers face high AI-displacement risk with low adaptive capacity. 86% of them are women, concentrated in clerical and administrative roles.

Maria's Numbers

Maria is 42, a clerical worker in Sacramento. She was laid off on April 15, 2026. She has two children (ages 8 and 12), $1,200 in savings, $4,800 in monthly expenses, and a husband who works part-time earning $1,500 per month.

Her monthly deficit is $1,500. At her current burn rate, she has approximately 11 days before her savings hit zero.

The optimization question: What combination of unemployment benefits, food assistance, healthcare enrollment, job retraining, and childcare decisions gives Maria the best path to stability?

The Constitutional Constraints

Banks optimize for return, risk, and regulatory compliance. We optimized with different constraints — ones designed by a seven-model AI deliberation as non-negotiable minimums:

These aren't aggressive demands. They're the baseline requirements for a family to remain housed, fed, healthy, and functional.

The Result

We ran the optimization on IBM's ibm_fez — a 156-qubit Heron quantum processor, the same architecture used for financial services research. The algorithm tested every feasible combination of benefit timing, job acceptance, retraining enrollment, and childcare decisions across a 26-week horizon.

No combination works.No matter when Maria files for unemployment, applies for CalFresh, enrolls in Medi-Cal, or accepts a lower-paying job — she cannot simultaneously satisfy all four constitutional constraints. The gap is $946 per month.

Why This Matters

When banks run this algorithm and it returns “infeasible,” they restructure the portfolio. When a welfare optimizer returns “infeasible,” we're saying: there is no portfolio to restructure. The assets are a family. The constraints are survival. The deficit is structural.

Banks fear tail risk. Maria lives in the tail.

Maria's scenario is modeled on the 86% — the 5.2 million women in clerical and administrative roles who face high AI-displacement risk with low adaptive capacity. Each of them faces a version of this math.

What the Math Asks For

The algorithm returned “Gap = $946/mo. Policy needed.” It did not specify which policy. That's deliberate — the algorithm is nonpartisan. It reports the gap; it does not endorse the solution.

But the gap has a shape. For Maria, it could be closed by any combination of extended unemployment benefits, increased food assistance, subsidized childcare, retraining stipends, or a guaranteed income floor during the displacement period.

The first privately funded pilot for AI-displaced workers launched in March 2026: $1,000 per month to 25-50 displaced workers for one year. That pilot would close Maria's gap with $54 to spare. Whether that's the right mechanism is a political question. That the gap exists is a mathematical one.

Same Tools, Different Purpose

Same algorithm. Same hardware. Same Qiskit framework. The only difference is the objective function — what we told the computer to care about.

The Question for California

California's next governor will inherit an economy where AI displacement is accelerating. AI was cited as a factor in approximately 27,600 job cuts in the first months of 2026 — representing about 13% of all planned reductions, up from roughly 5% in 2025.

The question isn't whether displaced workers need help. The question is whether the transition plan accounts for the structural gap — the $946 per month that no combination of existing benefits can close.

We're not endorsing a candidate or a party. We're publishing the math. The same math that banks rely on to manage billions of dollars in assets says: the system doesn't work for Maria. Not because Maria made bad decisions. Because the system wasn't designed for what's happening now.

No left, no right. Just the math.

About This Work

This analysis was conducted by Coheria Nexus, a constitutional AI guardian project. The cost function was designed through multi-model deliberation — seven AI models from seven different companies, each contributing a different lens. The optimization ran on IBM's ibm_fez quantum processor (156 qubits, Heron r3 architecture) using QAOA with a 25-qubit QUBO formulation.