What is an "advantaged solve" and why does it matter?

Pillar B · State as of: 2026-08-26

An advantaged solve is one problem instance where a quantum recipe measurably beat the strongest classical baseline its user could field — on the same instance, under the same budget, with the result measured, dated, and re-runnable. It is a unit of evidence, not a unit of billing: it counts verified victories one at a time, and it is equally honest about counting zero. As of August 2026, the number of advantaged solves verified in our own ledger is exactly that: zero.

Why does the field need a unit at all?

Because "quantum advantage" is a headline word, not a measurement. A decade-in-review survey (Khan, SSRN, Apr 2026) counts 450+ distinct quantum speedup results listed in the Quantum Algorithm Zoo — while the number of end-to-end, measured wins on a useful problem against a strong classical rival remains zero (the honest handful is short). Between those two numbers lives every press release in the industry. A unit forces the comparison to be specific: not "quantum beats classical at optimization", but "this recipe beat this baseline on this instance, on this date, under this budget — here are the artifacts."

The unit is deliberately small. It makes no claim about a problem class, a machine, or a vendor. One advantaged solve proves exactly one thing; a thousand of them, across instances and dates, would start to look like an industry.

What are the five requirements?

Each requirement closes a specific historical failure mode. Same instance closes the swap where the quantum side runs a friendlier input. Same budget closes the asymmetry where one side gets tuning time the other never saw — the failure that a weak classical baseline turns into a fake crossover. The strongest-baseline rule is the hard one, because the strongest rival sometimes does not exist yet when the claim ships: the dequantization wave (Tang, arXiv:1807.04271, 2018) retired a family of exponential machine-learning claims by building the classical rival that the original comparisons lacked. Measured-and-dated separates results from roadmaps. Re-runnable artifacts are what let anyone check the other four — radical reproducibility is not a virtue bolted on top of the unit; it is the unit's audit mechanism.

What happens to famous claims under this unit?

The recent record is what motivates the definition. Each row below is a real, honest piece of work whose headline outran the unit — none of this is an accusation, and each correction was itself published openly with re-runnable artifacts.

The infrastructure to make this unit routinely checkable is starting to exist: QOBLIB (Nature Computational Science, 2026) publishes 10 problem classes and 1,200+ exact instances with classical baselines, so "same instance, strong rival" stops being a promise and becomes a download.

How many advantaged solves exist today?

In our records: zero — including our own. The first sealed verdict of our ledger, V-0012, ran QAOA p=2 against CP-SAT on 20 sealed portfolio instances (n=12/16/20): CP-SAT reached proven optimum 20 times out of 20, QAOA landed 25–48% behind, and the quantum runs were noiseless simulations — a setup that favors the quantum side, declared as such — and it still lost every run (full context). That is a negative advantaged-solve count published under the same rules a positive would face. The symmetry is the point: a unit you only report when it flatters you is not a unit — it is marketing. This is one small problem class; it is not a universal verdict about anything.

If you want the reader-side version of these rules — how to evaluate someone else's claim in ten minutes — that is a separate checklist.

What we know / What we don't know

We know: the definition above is checkable, and every component targets a documented failure mode with at least one public episode behind it. We know our own count is zero, measured. We know the D-Wave dispute is open in both directions — the replication is published, the defense is published, and the unit refuses to adjudicate what has not been settled on shared instances.

We don't know: whether anyone, anywhere, has privately recorded a result that would pass all five requirements — absence from public record is not proof of absence. We don't know when the first verified advantaged solve on a commercially useful instance will happen; nothing in this definition predicts a date. And we don't know whether the industry will adopt any shared unit at all — no standards body currently owns one, and this page defines a measurement standard, not a market fact.

Rosetta Q publishes verdicts with reproducible raw data. This is educational content, not a product claim.

Sources:

• Tang — A quantum-inspired classical algorithm for recommendation systems, arXiv:1807.04271 (2018)
• Khan — Quantum Algorithms: A Decade in Review, SSRN (Apr 28, 2026)
• Kim et al. — Evidence for the utility of quantum computing before fault tolerance, Nature 618, 500 (2023)
• Tindall et al. — Efficient tensor network simulation of IBM's kicked Ising experiment, PRX Quantum 5, 010308 (2024)
• King et al. — Beyond-classical computation in quantum simulation, Science (Mar 2025)
• Tindall et al. — Dynamics of disordered quantum systems with two- and three-dimensional tensor networks, Science 392, 868 (Jul 21, 2026)
• D-Wave — "D-Wave's Quantum Supremacy Result Stands" (May 26, 2026)
• QOBLIB — Quantum Optimization Benchmarking Library, arXiv:2504.03832 (Nature Computational Science, 2026)
• Rosetta Q — Verdict V-0012 (2026)