Pillar D · State as of 2026-07-23

Run 001: what an honest quantum benchmark looks like (and why classical won)

We ran our first real fight: QAOA (quantum, simulated) vs OR-Tools CP-SAT (classical) on the same 12-asset portfolio instance, same time budget, fixed seed. CP-SAT reached the exact optimum in 0.113 s; QAOA landed 42.8% away in 45 s. Verdict: not yet — exactly as theory predicts at this scale. The run is sealed with a SHA-256 hash and archived in triplicate. This is entry 001 of the catalog.

→ Leer en español

State as of: 2026-07-23

What we ran

One instance of constrained portfolio optimization — pick 5 of 12 assets, balancing expected return against risk — encoded as a QUBO and handed to both contenders at once:

Classical side: Google’s OR-Tools CP-SAT, the kind of industrial solver a bank actually runs.
Quantum side: QAOA (depth p=2), simulated on CPU with PennyLane — 12 assets = 12 qubits.
The referee: the exact optimum, computed by brute force (2¹² = 4,096 candidates). At this size we can afford perfect truth, so nobody grades their own homework.

Same instance. Same 120-second budget. Seed 42 on everything. Library versions frozen and recorded.

The result

Measured result, sealed archive EXP-0012-001. Not an illustration.

CP-SAT found the provably optimal portfolio in a tenth of a second. QAOA, after 120 optimization steps and 2,000 measurement shots, delivered a portfolio 42.8% worse than the optimum — and took 400× longer doing it.

Why this was the expected result

Nobody serious predicts quantum advantage on a 12-variable problem, and we didn’t either — the archive records our hypothesis verbatim: “at 12 assets NO quantum advantage is expected; this run fixes the protocol’s baseline.” Three reasons the classical side dominates here:

The problem is tiny. 4,096 candidate portfolios is nothing; CP-SAT proves optimality almost instantly.
QAOA at depth p=2 is a shallow heuristic. Its approximation quality grows with circuit depth — and depth is exactly what’s scarce on both simulators and real hardware.
The baseline is strong on purpose. Beating a weakened classical solver is the most common sin in quantum benchmarking. A win against a weak baseline is not a win.

Then why run it at all?

Because a verification ledger earns trust with its no’s before its yes’s. This run establishes three assets:

The protocol, live. Same instance, same budget, exact-optimum referee, fixed seeds, frozen versions — now demonstrated end to end, not described.
The baseline of the crossover curve. Every future run — 16 assets, 20, deeper circuits, real QPUs — gets compared against this point. Where (and whether) the gap closes is the product.
Radical reproducibility. The archive records seed, instance parameters and library versions. Run the published harness with those values and you should land on our numbers. If you can’t, tell us — publicly.

Integrity of this entry

The full run lives in a JSON archive sealed with sha256:d0a207d6…bce6c, stored simultaneously in three places (GitHub, Codeberg, and our database) that reference each other. If any copy’s hash differs, that copy is invalid. The archive answers six questions — what, how, when, where, why, who — so this entry can be audited without trusting this post.

What we don’t know

Whether, and where, QAOA-family recipes cross over CP-SAT on portfolio problems as instances grow — that is precisely what the coming runs measure. We also don’t yet know how these CPU-simulator results transfer to noisy hardware. We will publish both, whichever way they land.

Measured content from sealed archive EXP-0012-001. Rosetta Quantum publishes verdicts with reproducible raw data — including the ones quantum loses.

Sources:

· Sealed archive EXP-0012-001 (sha256:d0a207d6…bce6c), RosettaQ evidence ledger" }