Who submits the classical rival on the Quantum Advantage Tracker?
The Quantum Advantage Tracker, launched by IBM and Algorithmiq in November 2025, does something the field badly needed: it puts quantum results and classical rivals on the same instance, with circuits and data attached, where anyone can look. We use it. This note is about one thing we found while looking, and it is not a criticism of the tracker — it is a measurement of who is showing up.
On 10 September 2026 we downloaded the three submissions.json files from the tracker's own repository and counted. Not the homepage summary: the data.
What we counted
Forty-seven entries across the three registries. Seventeen of them are quantum runs; the rest are classical methods submitted as rivals or baselines. Twelve instances have both a quantum arm and at least one classical challenger — those are the only ones where a comparison exists at all.
In 8 of those 12, every classical rival shares a submitting institution with the team that ran the quantum experiment.
The concentration is not spread evenly, and where it clusters is the interesting part:
| Registry | Instances with both arms | Only same institutions | With an outside challenger |
|---|---|---|---|
| Observable estimations | 5 | 5 | 0 |
| Variational problems | 4 | 3 | 1 |
| Classically verifiable problems | 3 | 0 | 3 |
Observable estimations is the registry where expectation values arrive with error bars — the format in which advantage gets announced. No third party has ever submitted a classical method there. Classically verifiable problems is where the answer can be checked from outside without anyone's permission. Every instance there is contested by an outsider.
Outside scrutiny appears where a result can be verified without permission, and disappears where you have to trust the party doing the measuring.
The clearest case
The tracker publishes institutions for all 47 entries and author names for 16. On the instance guided_sparse_ground_state_problem_49Q_v1 the names are enough to settle it: the quantum entry (SKQD, from IBM, RIKEN and the University of Chicago) and all five classical rivals — DMRG, Truncated Arnoldi, ASCI, CIPSI and HCI — carry the identical author list, person by person. It is not shared employers. It is one team competing against itself, six times.
Four cases that cut the other way
We are naming these because the finding is weaker without them, and because a claim that only survives when you omit the counter-examples is not a finding.
- Moderna wins on a laptop CPU. On
peaked_circuit_P9_Hqap_56x1917, the fastest classical entry is MPO + Unswapping (CPU), submitted by Moderna — unconnected to the consortium that ran the quantum arm. It beats that consortium's own classical entry, which is marked as extrapolated. - Wisconsin–Madison beats a quantum arm. On
4fe_4s_54e_36o, TrimCI from the University of Wisconsin–Madison reaches a lower energy than the quantum SQD run from IBM and RIKEN. - Two teams published the classical method that beats them. On
2fe_2s_30e_20oandn2_3-00_14e_28o, Qunova Computing and RIKEN submitted both the quantum HI-VQE and the classical HCI that outperforms it. - IBM against IBM. On both
peaked_circuit_heavy_hexinstances, the fastest classical entry is IBM's, competing with IBM's quantum arm.
What this is not
It is not evidence that anyone shaped a result. Submitting the strongest classical rival to your own quantum claim is the honest thing to do, and several groups here did exactly that — including against themselves. The measurement is about who else showed up, and the answer for the loudest registry is: nobody, yet.
It is also not the same question as reproduction. Across all 47 entries there is no instance where two quantum arms come from disjoint groups — that is, nobody has re-run another team's quantum experiment and confirmed its headline result. Refuting a claim, verifying an output and competing against it with a classical method are three different acts, and all three have happened. Reproduction has not.
The distinction matters in a case that is often cited as if it settled the question: in the doped Clifford episode, a team at the Singapore University of Technology and Design computed the exact amplitudes of the 2,051 outputs IBM had published. That is independent verification of the output, and it is real. It is not reproduction of the experiment, and it is not confirmation of the advantage — what narrowed was the time gap, while the certificate held.
Check it yourself
Everything above comes from three files. Download data/observable-estimations/submissions.json, data/classically-verifiable-problems/submissions.json and data/variational-problems/submissions.json from the tracker's repository, group the entries by circuit or hamiltonian, and compare the institutions field of each classical entry against the quantum one on the same instance. It takes about twenty lines of code. If your count differs from ours, we want to know.
We will re-run this on the next edition. The number worth watching is not 8 of 12 — it is whether the first outside classical submission ever lands in observable estimations.