Constrained portfolio compression: does QAOA beat the classical champion?
Class: portfolio optimization · Quantum side: QAOA p=2 (PennyLane, exact simulator) · Classical side: OR-Tools CP-SAT · Same instance, same budget, exact optimum as referee
CLAIM"QAOA can match or beat classical solvers on constrained portfolio selection at small n." — internal hypothesis, pre-registered before any run (PR-ITER2-001)
In plain termsWe put a quantum algorithm and a classical one on the same investment problem, same time, same machine. The classical one found the perfect answer in all twenty tests; the quantum one always fell short — between 10% and 87% depending on the seed — and the gap does not shrink consistently as the problem grows. The question stays open for one measured reason: the classical solver's time to prove it found the best answer grew about 25× for every four assets added.
The five layers
0 · INTEGRITY
Harness harness_v0.py published with the record; instances regenerated from declared seeds; both sides run on the same machine under a 120 s budget.
1 · VERDICT
Criterion sealed before any run: advantage = QAOA reaching the exact optimum at least as often as CP-SAT within budget. Result: never.
2 · MAGNITUDE
Mean gaps 48.2 / 25.1 / 41.1 % with seed dispersion 10–23 points; every one of the 20 individual gaps is drawn below and recomputed from the sealed runs.
3 · ROBUSTNESS
Seed ranges overlap across n=12/16/20; the verdict does not depend on any single seed or size. QAOA never wins against a strong baseline; whether it wins against weak ones was not the question.
4 · TRIGGERS
Measured side note: CP-SAT's optimality-proof time grew ~25× per +4 assets (0.05 s → 1.3 s → 29 s). If that curve continues, the budget-parity fight at n ≥ 28 is a different fight.
5 · YOUR UNITS
Set with the client: the gap translated into the rebalancing book, frequency and cost of suboptimality — supplied by the client or marked as assumption.
"CP-SAT alcanzo el optimo exacto probado en los 20 runs sellados (n=12: 8 seeds, n=16: 8, n=20: 4). Brechas medias de QAOA p=2 al optimo: 48.2% (sigma 22.5), 25.1% (sigma 11.1), 41.1% (sigma 14.1). Sin tendencia de tamano defendible: los rangos entre semillas se solapan."— verbatim from the sealed verdict file, unedited (ASCII, no accents in the original)
What this page is. Page 1 of every Rosetta Quantum deliverable. If it does not stand on its own, the report is wrong. A full Diligence Report adds 25–40 pages behind it: instances, harness, every run, and the economics in the client's units.
Who paid. Nobody: this is our own experiment, published under the same rules we apply to everyone. Judgment is paid for by those who rely on it — never by the vendors being evaluated.
"Not yet" and "no advantage measured" are measurements under the pre-registered criterion, not allegations. Each source states its own caveats.
Rosetta Quantum is a brand of Blue Tuna SpA · hello@rosettaquantum.com · Verify: download the sealed file and recompute sha256 · Methodology v1.0 · juez-v1